Hugging Face's logo

Datasets:
paintedwolfcode
/
bialy-dataset

Modalities:
Text
Formats:
json
Size:
1K - 10K
Tags:
Synthetic
coding-agents
tool-selection
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
paintedwolf-ai commited on
Commit
48f76bc
·
verified ·
1 Parent(s): 24918f7

Release open1-b7g-e4

Browse files
Files changed (21) hide show
  1. FILES.json +18 -2322
  2. NOTICE +1 -1
  3. PROVENANCE.json +5 -427
  4. README.md +11 -231
  5. SHA256SUMS +20 -261
  6. TRAINING.md +13 -71
  7. corpora/independent.json +71 -33
  8. corpus.json +0 -0
  9. holdout.jsonl +0 -0
  10. train.jsonl +2 -2
  11. training/turn-load/guide_audit.py +66 -0
  12. training/turn-load/relabel_guides.py +75 -0
  13. training/turn-load/scripts/decide/README.md +358 -0
  14. training/turn-load/scripts/decide/corpus.py +11 -4
  15. training/turn-load/scripts/decide/parity_probe.py +9 -22
  16. training/turn-load/scripts/decide/rerank/README.md +173 -0
  17. training/turn-load/scripts/decide/rerank/train_rerank.py +2 -2
  18. training/turn-load/scripts/decide/train.py +39 -25
  19. training/turn-load/train.sh +61 -0
  20. training/turn-load/turn_probe.py +103 -0
  21. val.jsonl +0 -0
FILES.json CHANGED
@@ -1,2333 +1,29 @@
1
  [
2
  {
3
  "path": "train.jsonl",
4
- "source_sha256": "2ec8ce650fd19b0e22bbc6b1673be699165fa258e98bbf5f963e6c9354ba80bd",
5
- "sha256": "ec3a0008eb903cd455912ce094a1e53884498a77ef76a20fa72805ab7b9aaeb5",
6
- "source_bytes": 27830700,
7
- "bytes": 27801342,
8
- "replacements": {
9
- "archive_path": 338
10
- },
11
- "rows": 2746
12
  },
13
  {
14
  "path": "val.jsonl",
15
- "source_sha256": "67c4c86e3385f0241048f2633d97a4aad78a53a3b7a540592937a20eba669fd3",
16
- "sha256": "a2f446c6b3b34b2dec1b70a04836ebc55e7951bcc4bcca0556019cb030dcda30",
17
- "source_bytes": 5001848,
18
- "bytes": 4998899,
19
- "replacements": {
20
- "archive_path": 34
21
- },
22
- "rows": 489
23
  },
24
  {
25
  "path": "holdout.jsonl",
26
- "source_sha256": "e33344a6d0dda501dcab46b08317205b7e508a71f4f0781b1c257c32d400ee22",
27
- "sha256": "f50ec8ba164e5b01536b02c5d62c81285dee3732822b1381e39a69b869ed80c9",
28
- "source_bytes": 7549762,
29
- "bytes": 7542702,
30
- "replacements": {
31
- "archive_path": 80
32
- },
33
- "rows": 739
34
- },
35
- {
36
- "path": "corpus.json",
37
- "source_sha256": "51e3a65258fc34581bc8cfaae0f7c97f444f0a50f3dda906c77b6554fd5c66d3",
38
- "sha256": "51e3a65258fc34581bc8cfaae0f7c97f444f0a50f3dda906c77b6554fd5c66d3",
39
- "source_bytes": 117757,
40
- "bytes": 117757,
41
- "replacements": {}
42
- },
43
- {
44
- "path": "corpora/independent.json",
45
- "source_sha256": "c77abd8b40965c05cc43a7545e8c2db2405bdcfc18a73bad254632d523b67bcd",
46
- "sha256": "c77abd8b40965c05cc43a7545e8c2db2405bdcfc18a73bad254632d523b67bcd",
47
- "source_bytes": 136130,
48
- "bytes": 136130,
49
- "replacements": {}
50
- },
51
- {
52
- "path": "unit-rank/selection.jsonl",
53
- "source_sha256": "cfab42d7f33424f219fe9d90a9f94ff1d88c08c3a25a8969956d384b3e8306a5",
54
- "sha256": "20a3ebd2501f6b4ad092827eca2d9de79e477847d0ff0718f13adc87d7854ed9",
55
- "source_bytes": 7589861,
56
- "bytes": 7586912,
57
- "replacements": {
58
- "archive_path": 34
59
- },
60
- "rows": 1024
61
- },
62
- {
63
- "path": "unit-rank/generated-train.jsonl",
64
- "source_sha256": "de4c3becb1dbeeaa208cba89cf229a8150781deabe0034f227256c0d5eb57503",
65
- "sha256": "de4c3becb1dbeeaa208cba89cf229a8150781deabe0034f227256c0d5eb57503",
66
- "source_bytes": 12724155,
67
- "bytes": 12724155,
68
- "replacements": {},
69
- "rows": 2634
70
- },
71
- {
72
- "path": "unit-rank/train.jsonl",
73
- "source_sha256": "a82ec5029e5e0d3323810314a31855cb8db7f0d3ed7145d06503bb881fd6f06b",
74
- "sha256": "5c857914f92211be68a6c9b8087c5589155c48ef0182c986484c5588f43b999f",
75
- "source_bytes": 32885301,
76
- "bytes": 32855943,
77
- "replacements": {
78
- "archive_path": 338
79
- },
80
- "rows": 3792
81
- },
82
- {
83
- "path": "unit-rank/historical-acceptance.jsonl",
84
- "source_sha256": "5fda6375ca6d6ffeb08fd8d20e7c2342c5051101ab77d52231178e892093f745",
85
- "sha256": "5fda6375ca6d6ffeb08fd8d20e7c2342c5051101ab77d52231178e892093f745",
86
- "source_bytes": 4823488,
87
- "bytes": 4823488,
88
- "replacements": {},
89
- "rows": 789
90
- },
91
- {
92
- "path": "evidence/preparation.json",
93
- "source_sha256": "80bb261c8433594f12508a60f7a859df01b526e8a7ff47993e817256a72b4b4b",
94
- "sha256": "80bb261c8433594f12508a60f7a859df01b526e8a7ff47993e817256a72b4b4b",
95
- "source_bytes": 205,
96
- "bytes": 205,
97
- "replacements": {}
98
- },
99
- {
100
- "path": "tasks.jsonl",
101
- "source_sha256": "23a4abda8e51e7fc821615b5a83dc250d115d702cad19da4cf68863e260b1d43",
102
- "sha256": "23a4abda8e51e7fc821615b5a83dc250d115d702cad19da4cf68863e260b1d43",
103
- "source_bytes": 2342008,
104
- "bytes": 2342008,
105
- "replacements": {},
106
- "rows": 3024
107
- },
108
- {
109
- "path": "sessions/final-collected-rows.jsonl",
110
- "source_sha256": "b886be4e60b94a9ae27a6e36fe60ad2aafcf2ac8d574170dffd79fded77ac152",
111
- "sha256": "b886be4e60b94a9ae27a6e36fe60ad2aafcf2ac8d574170dffd79fded77ac152",
112
- "source_bytes": 12429682,
113
- "bytes": 12429682,
114
- "replacements": {},
115
- "rows": 3974
116
- },
117
- {
118
- "path": "sessions/live-snapshot-rows.jsonl",
119
- "source_sha256": "6d8af5d81cbd81874f5a4d6ce1df9dad1eba9cd66e8ee534f01057a1cb41eae5",
120
- "sha256": "6d8af5d81cbd81874f5a4d6ce1df9dad1eba9cd66e8ee534f01057a1cb41eae5",
121
- "source_bytes": 1232080,
122
- "bytes": 1232080,
123
- "replacements": {},
124
- "rows": 388
125
- },
126
- {
127
- "path": "sessions/combined-snapshot-rows.jsonl",
128
- "source_sha256": "65752a429f324f36e28ba17e721d3fe2853b010b6a44cdb921d8a18592a2f62c",
129
- "sha256": "65752a429f324f36e28ba17e721d3fe2853b010b6a44cdb921d8a18592a2f62c",
130
- "source_bytes": 11677948,
131
- "bytes": 11677948,
132
- "replacements": {},
133
- "rows": 3741
134
- },
135
- {
136
- "path": "sessions/later-rows.jsonl",
137
- "source_sha256": "a24c3a26bcee3f5e9c1af79c447ea38866e2ef2d4d2a03da9452f0e0bb92f240",
138
- "sha256": "a24c3a26bcee3f5e9c1af79c447ea38866e2ef2d4d2a03da9452f0e0bb92f240",
139
- "source_bytes": 756002,
140
- "bytes": 756002,
141
- "replacements": {},
142
- "rows": 233
143
- },
144
- {
145
- "path": "historical/open1-judged.jsonl",
146
- "source_sha256": "4923f345359a0df64513d6cd2f689483df5b1a3081734760c2fbaf9235df7fe0",
147
- "sha256": "4923f345359a0df64513d6cd2f689483df5b1a3081734760c2fbaf9235df7fe0",
148
- "source_bytes": 22247608,
149
- "bytes": 22247608,
150
- "replacements": {},
151
- "rows": 3974
152
- },
153
- {
154
- "path": "historical/train.jsonl",
155
- "source_sha256": "0fea3a5ab5eee75148c57252b062f1bbfa934bfbe561b3c1a00ac03f8fec137d",
156
- "sha256": "0fea3a5ab5eee75148c57252b062f1bbfa934bfbe561b3c1a00ac03f8fec137d",
157
- "source_bytes": 15338794,
158
- "bytes": 15338794,
159
- "replacements": {},
160
- "rows": 2746
161
- },
162
- {
163
- "path": "historical/val.jsonl",
164
- "source_sha256": "d0c39770ec9776849c42adc9e4c70b64c2463574b8e306831694279e68ca1bd6",
165
- "sha256": "d0c39770ec9776849c42adc9e4c70b64c2463574b8e306831694279e68ca1bd6",
166
- "source_bytes": 2744073,
167
- "bytes": 2744073,
168
- "replacements": {},
169
- "rows": 489
170
- },
171
- {
172
- "path": "historical/holdout.jsonl",
173
- "source_sha256": "3cac53164d5a4c4f81e3fdce6001b08c8cf13e2a739ad08558cd2baa9b227bef",
174
- "sha256": "3cac53164d5a4c4f81e3fdce6001b08c8cf13e2a739ad08558cd2baa9b227bef",
175
- "source_bytes": 4164741,
176
- "bytes": 4164741,
177
- "replacements": {},
178
- "rows": 739
179
- },
180
- {
181
- "path": "agreement.jsonl",
182
- "source_sha256": "7e09a014dd94f32d95196a3531e046c1904d526a593b670f448f1d87168a44ff",
183
- "sha256": "7e09a014dd94f32d95196a3531e046c1904d526a593b670f448f1d87168a44ff",
184
- "source_bytes": 9316780,
185
- "bytes": 9316780,
186
- "replacements": {},
187
- "rows": 41091
188
- },
189
- {
190
- "path": "corpora/original-open1.json",
191
- "source_sha256": "51e3a65258fc34581bc8cfaae0f7c97f444f0a50f3dda906c77b6554fd5c66d3",
192
- "sha256": "51e3a65258fc34581bc8cfaae0f7c97f444f0a50f3dda906c77b6554fd5c66d3",
193
- "source_bytes": 117757,
194
- "bytes": 117757,
195
- "replacements": {}
196
- },
197
- {
198
- "path": "row.schema.json",
199
- "source_sha256": "260f26ccdd53437f1d01b269d9f510dde1b497b9b64a45a7807d83231a2a7bb5",
200
- "sha256": "260f26ccdd53437f1d01b269d9f510dde1b497b9b64a45a7807d83231a2a7bb5",
201
- "source_bytes": 6241,
202
- "bytes": 6241,
203
- "replacements": {}
204
- },
205
- {
206
- "path": "coderank/pairs/docs-axios.jsonl",
207
- "source_sha256": "998c6d5eedc8c7443f3de5650007a8cb06d43cc4399d246af43d9561c47301e4",
208
- "sha256": "998c6d5eedc8c7443f3de5650007a8cb06d43cc4399d246af43d9561c47301e4",
209
- "source_bytes": 2339,
210
- "bytes": 2339,
211
- "replacements": {},
212
- "rows": 9
213
- },
214
- {
215
- "path": "coderank/pairs/docs-axum.jsonl",
216
- "source_sha256": "530ab0dbb7109c750d2bd46e09caa143e13261442f343f7bc55451b26942f8f3",
217
- "sha256": "530ab0dbb7109c750d2bd46e09caa143e13261442f343f7bc55451b26942f8f3",
218
- "source_bytes": 15143,
219
- "bytes": 15143,
220
- "replacements": {},
221
- "rows": 64
222
- },
223
- {
224
- "path": "coderank/pairs/docs-cobra.jsonl",
225
- "source_sha256": "be739e8b08ec8a9eb37f84ac1e50e7052b90855f60df57f89a45d2c27b05b62c",
226
- "sha256": "be739e8b08ec8a9eb37f84ac1e50e7052b90855f60df57f89a45d2c27b05b62c",
227
- "source_bytes": 7844,
228
- "bytes": 7844,
229
- "replacements": {},
230
- "rows": 33
231
- },
232
- {
233
- "path": "coderank/pairs/docs-express.jsonl",
234
- "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
235
- "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
236
- "source_bytes": 0,
237
- "bytes": 0,
238
- "replacements": {},
239
- "rows": 0
240
- },
241
- {
242
- "path": "coderank/pairs/docs-fd.jsonl",
243
- "source_sha256": "c9277637cf12e41c3bcbdb248d77b13ce8b3e4703dd124ed97b0d259a1566466",
244
- "sha256": "c9277637cf12e41c3bcbdb248d77b13ce8b3e4703dd124ed97b0d259a1566466",
245
- "source_bytes": 3336,
246
- "bytes": 3336,
247
- "replacements": {},
248
- "rows": 16
249
- },
250
- {
251
- "path": "coderank/pairs/docs-flask.jsonl",
252
- "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
253
- "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
254
- "source_bytes": 0,
255
- "bytes": 0,
256
- "replacements": {},
257
- "rows": 0
258
- },
259
- {
260
- "path": "coderank/pairs/docs-gin.jsonl",
261
- "source_sha256": "77f196de917b9056f37c51aa506dd99f73e627d306633864bcabf48e19e082bf",
262
- "sha256": "77f196de917b9056f37c51aa506dd99f73e627d306633864bcabf48e19e082bf",
263
- "source_bytes": 15769,
264
- "bytes": 15769,
265
- "replacements": {},
266
- "rows": 72
267
- },
268
- {
269
- "path": "coderank/pairs/docs-gson.jsonl",
270
- "source_sha256": "089466b11db11704ae4bbcf51b16eaee28b152f3374350fa2abc1df51b3c1548",
271
- "sha256": "089466b11db11704ae4bbcf51b16eaee28b152f3374350fa2abc1df51b3c1548",
272
- "source_bytes": 21239,
273
- "bytes": 21239,
274
- "replacements": {},
275
- "rows": 77
276
- },
277
- {
278
- "path": "coderank/pairs/docs-httpx.jsonl",
279
- "source_sha256": "4585210652c4ae8920fb3627deb6762de7c9b3fc91e482b7a3e738d98e3f07d1",
280
- "sha256": "4585210652c4ae8920fb3627deb6762de7c9b3fc91e482b7a3e738d98e3f07d1",
281
- "source_bytes": 228,
282
- "bytes": 228,
283
- "replacements": {},
284
- "rows": 1
285
- },
286
- {
287
- "path": "coderank/pairs/docs-okhttp.jsonl",
288
- "source_sha256": "6e3b828499f84eeef1a16d3563f29c6ef3e45b9d1248ee03a23e4a54ed5ddf4a",
289
- "sha256": "6e3b828499f84eeef1a16d3563f29c6ef3e45b9d1248ee03a23e4a54ed5ddf4a",
290
- "source_bytes": 53524,
291
- "bytes": 53524,
292
- "replacements": {},
293
- "rows": 188
294
- },
295
- {
296
- "path": "coderank/pairs/docs-requests.jsonl",
297
- "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
298
- "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
299
- "source_bytes": 0,
300
- "bytes": 0,
301
- "replacements": {},
302
- "rows": 0
303
- },
304
- {
305
- "path": "coderank/pairs/docs-ripgrep.jsonl",
306
- "source_sha256": "f4aed9ff9ba9b581b4f8c852821d6e3f14e2291da3c895da01e2bf33aa7a886b",
307
- "sha256": "f4aed9ff9ba9b581b4f8c852821d6e3f14e2291da3c895da01e2bf33aa7a886b",
308
- "source_bytes": 60417,
309
- "bytes": 60417,
310
- "replacements": {},
311
- "rows": 244
312
- },
313
- {
314
- "path": "coderank/pairs/docs-sinatra.jsonl",
315
- "source_sha256": "e04687b8333e1d8aceca7f9599d0092a56c6f2d32ba4a0f72e6b8f1925bb4047",
316
- "sha256": "e04687b8333e1d8aceca7f9599d0092a56c6f2d32ba4a0f72e6b8f1925bb4047",
317
- "source_bytes": 7364,
318
- "bytes": 7364,
319
- "replacements": {},
320
- "rows": 29
321
- },
322
- {
323
- "path": "coderank/pairs/docs-slim.jsonl",
324
- "source_sha256": "b9178f3efa0df34e16cf2c32e9424ed55e9aa9a4306368b3719ea40f4b1ec551",
325
- "sha256": "b9178f3efa0df34e16cf2c32e9424ed55e9aa9a4306368b3719ea40f4b1ec551",
326
- "source_bytes": 2687,
327
- "bytes": 2687,
328
- "replacements": {},
329
- "rows": 10
330
- },
331
- {
332
- "path": "coderank/pairs/docs-thor.jsonl",
333
- "source_sha256": "4f40152e46b8ca5e5d6540e567a36ef1f94493130f61a05ae7e8a26dc5c47427",
334
- "sha256": "4f40152e46b8ca5e5d6540e567a36ef1f94493130f61a05ae7e8a26dc5c47427",
335
- "source_bytes": 5454,
336
- "bytes": 5454,
337
- "replacements": {},
338
- "rows": 24
339
- },
340
- {
341
- "path": "coderank/pairs/docs-zod.jsonl",
342
- "source_sha256": "cc412c5833ee9e28f4b92058c4605fd108c0fda7196317a3c5c0843c4f8ec302",
343
- "sha256": "cc412c5833ee9e28f4b92058c4605fd108c0fda7196317a3c5c0843c4f8ec302",
344
- "source_bytes": 50223,
345
- "bytes": 50223,
346
- "replacements": {},
347
- "rows": 189
348
- },
349
- {
350
- "path": "coderank/pairs/model-axios.jsonl",
351
- "source_sha256": "8f12d3deb77a884f577ffa10bbb4628260c71c5e2be8c1ace6cf8a5f4bd10503",
352
- "sha256": "8f12d3deb77a884f577ffa10bbb4628260c71c5e2be8c1ace6cf8a5f4bd10503",
353
- "source_bytes": 204992,
354
- "bytes": 204992,
355
- "replacements": {},
356
- "rows": 588
357
- },
358
- {
359
- "path": "coderank/pairs/model-axum.jsonl",
360
- "source_sha256": "7fc85185b21b26ade374034ee55a3b62b5c9abf7b0ec8f33106957292d5ba18f",
361
- "sha256": "7fc85185b21b26ade374034ee55a3b62b5c9abf7b0ec8f33106957292d5ba18f",
362
- "source_bytes": 279483,
363
- "bytes": 279483,
364
- "replacements": {},
365
- "rows": 794
366
- },
367
- {
368
- "path": "coderank/pairs/model-cobra.jsonl",
369
- "source_sha256": "767cbf2ca31e728aa6dde8080cb8eddff6273688d213adc46d5bbad40d695664",
370
- "sha256": "767cbf2ca31e728aa6dde8080cb8eddff6273688d213adc46d5bbad40d695664",
371
- "source_bytes": 97567,
372
- "bytes": 97567,
373
- "replacements": {},
374
- "rows": 287
375
- },
376
- {
377
- "path": "coderank/pairs/model-express.jsonl",
378
- "source_sha256": "f7d7e19f2d6f06b8eca2c0223f3c43dce3eca1370ba6a2517a6b0801501d7e8c",
379
- "sha256": "f7d7e19f2d6f06b8eca2c0223f3c43dce3eca1370ba6a2517a6b0801501d7e8c",
380
- "source_bytes": 33445,
381
- "bytes": 33445,
382
- "replacements": {},
383
- "rows": 105
384
- },
385
- {
386
- "path": "coderank/pairs/model-fd.jsonl",
387
- "source_sha256": "0f486dfa2d340155f96b9c3e1ef8060dadee0a9c6d415c96c5bf13cf8abf7ca9",
388
- "sha256": "0f486dfa2d340155f96b9c3e1ef8060dadee0a9c6d415c96c5bf13cf8abf7ca9",
389
- "source_bytes": 79153,
390
- "bytes": 79153,
391
- "replacements": {},
392
- "rows": 240
393
- },
394
- {
395
- "path": "coderank/pairs/model-flask.jsonl",
396
- "source_sha256": "25bedb286f7b2dbb5025b93f6e11f3de3f9f487027d160fc1f14211c89c5dc91",
397
- "sha256": "25bedb286f7b2dbb5025b93f6e11f3de3f9f487027d160fc1f14211c89c5dc91",
398
- "source_bytes": 258357,
399
- "bytes": 258357,
400
- "replacements": {},
401
- "rows": 787
402
- },
403
- {
404
- "path": "coderank/pairs/model-gin.jsonl",
405
- "source_sha256": "1451daabf55fc3a83544fcc3e28d64dc91f361d15e3955852d2e148f26184e11",
406
- "sha256": "1451daabf55fc3a83544fcc3e28d64dc91f361d15e3955852d2e148f26184e11",
407
- "source_bytes": 227039,
408
- "bytes": 227039,
409
- "replacements": {},
410
- "rows": 702
411
- },
412
- {
413
- "path": "coderank/pairs/model-gson.jsonl",
414
- "source_sha256": "e674e208bc303e7fa89de0b5454f837e13d82b07623b6e0db34754669f6382ad",
415
- "sha256": "e674e208bc303e7fa89de0b5454f837e13d82b07623b6e0db34754669f6382ad",
416
- "source_bytes": 243841,
417
- "bytes": 243841,
418
- "replacements": {},
419
- "rows": 619
420
- },
421
- {
422
- "path": "coderank/pairs/model-httpx.jsonl",
423
- "source_sha256": "99f4aef319a2926698674859265f8a13f8c4adc7b754077421f53a5441891a47",
424
- "sha256": "99f4aef319a2926698674859265f8a13f8c4adc7b754077421f53a5441891a47",
425
- "source_bytes": 187238,
426
- "bytes": 187238,
427
- "replacements": {},
428
- "rows": 547
429
- },
430
- {
431
- "path": "coderank/pairs/model-okhttp.jsonl",
432
- "source_sha256": "9b0a61541a15586f85e1f810a5d1ee7c55e78cc50e14050370be9bb0ef7db809",
433
- "sha256": "9b0a61541a15586f85e1f810a5d1ee7c55e78cc50e14050370be9bb0ef7db809",
434
- "source_bytes": 236411,
435
- "bytes": 236411,
436
- "replacements": {},
437
- "rows": 615
438
- },
439
- {
440
- "path": "coderank/pairs/model-requests.jsonl",
441
- "source_sha256": "d452b47569a1721d17323a2a7b33fde102f63abaac9176ae47a1d48963dd907a",
442
- "sha256": "d452b47569a1721d17323a2a7b33fde102f63abaac9176ae47a1d48963dd907a",
443
- "source_bytes": 135629,
444
- "bytes": 135629,
445
- "replacements": {},
446
- "rows": 406
447
- },
448
- {
449
- "path": "coderank/pairs/model-ripgrep.jsonl",
450
- "source_sha256": "aecb05e06219fa1274cc76214864bcb041acefa0c6bf3b3c60c510bdb9eeefd6",
451
- "sha256": "aecb05e06219fa1274cc76214864bcb041acefa0c6bf3b3c60c510bdb9eeefd6",
452
- "source_bytes": 327129,
453
- "bytes": 327129,
454
- "replacements": {},
455
- "rows": 939
456
- },
457
- {
458
- "path": "coderank/pairs/model-sinatra.jsonl",
459
- "source_sha256": "804e473f4d7d940241ed55d18b736df91501a15b0df9de566a89848d04ce5a6d",
460
- "sha256": "804e473f4d7d940241ed55d18b736df91501a15b0df9de566a89848d04ce5a6d",
461
- "source_bytes": 181707,
462
- "bytes": 181707,
463
- "replacements": {},
464
- "rows": 531
465
- },
466
- {
467
- "path": "coderank/pairs/model-slim.jsonl",
468
- "source_sha256": "06e31042ffca6f1105f182d0c652efe98b9dc4fbb33b62cdef7d0d3acbb71f69",
469
- "sha256": "06e31042ffca6f1105f182d0c652efe98b9dc4fbb33b62cdef7d0d3acbb71f69",
470
- "source_bytes": 143660,
471
- "bytes": 143660,
472
- "replacements": {},
473
- "rows": 396
474
- },
475
- {
476
- "path": "coderank/pairs/model-thor.jsonl",
477
- "source_sha256": "61612faa4d87a194d40ccf95a96665d10cf986ae5eb0153f4851c06b4ee94c9c",
478
- "sha256": "61612faa4d87a194d40ccf95a96665d10cf986ae5eb0153f4851c06b4ee94c9c",
479
- "source_bytes": 167113,
480
- "bytes": 167113,
481
- "replacements": {},
482
- "rows": 511
483
- },
484
- {
485
- "path": "coderank/pairs/model-zod.jsonl",
486
- "source_sha256": "66abd50a893a931bf9c00a774d07e81e122fad83b025da2722880c20a770e644",
487
- "sha256": "66abd50a893a931bf9c00a774d07e81e122fad83b025da2722880c20a770e644",
488
- "source_bytes": 319483,
489
- "bytes": 319483,
490
- "replacements": {},
491
- "rows": 939
492
- },
493
- {
494
- "path": "coderank/units/axios.jsonl",
495
- "source_sha256": "9f889775e9a19377d060bacc09f7a11bb09ed50acbf523d425e392fbcdcd5f99",
496
- "sha256": "9f889775e9a19377d060bacc09f7a11bb09ed50acbf523d425e392fbcdcd5f99",
497
- "source_bytes": 454589,
498
- "bytes": 454589,
499
- "replacements": {},
500
- "rows": 1068
501
- },
502
- {
503
- "path": "coderank/units/axum.jsonl",
504
- "source_sha256": "f84ef9bd37d373f3117aaa7c256b581ab8670697030a2c145cd9b4680a96a380",
505
- "sha256": "f84ef9bd37d373f3117aaa7c256b581ab8670697030a2c145cd9b4680a96a380",
506
- "source_bytes": 1230497,
507
- "bytes": 1230497,
508
- "replacements": {},
509
- "rows": 2691
510
- },
511
- {
512
- "path": "coderank/units/cobra.jsonl",
513
- "source_sha256": "2932a270c2b5b76d7a4dd699fa569945db4bc0b4bc549110545dcb684f95485e",
514
- "sha256": "2932a270c2b5b76d7a4dd699fa569945db4bc0b4bc549110545dcb684f95485e",
515
- "source_bytes": 299052,
516
- "bytes": 299052,
517
- "replacements": {},
518
- "rows": 553
519
- },
520
- {
521
- "path": "coderank/units/express.jsonl",
522
- "source_sha256": "7ce8be3a2bea8553a163924cb5577279b4ff0d2a00389fd747568f4dd27e2edf",
523
- "sha256": "7ce8be3a2bea8553a163924cb5577279b4ff0d2a00389fd747568f4dd27e2edf",
524
- "source_bytes": 41609,
525
- "bytes": 41609,
526
- "replacements": {},
527
- "rows": 131
528
- },
529
- {
530
- "path": "coderank/units/fd.jsonl",
531
- "source_sha256": "e2e4cb29127163aea036ee1139d88225f09858597507553063b05f1e2a91cdb7",
532
- "sha256": "e2e4cb29127163aea036ee1139d88225f09858597507553063b05f1e2a91cdb7",
533
- "source_bytes": 201692,
534
- "bytes": 201692,
535
- "replacements": {},
536
- "rows": 404
537
- },
538
- {
539
- "path": "coderank/units/flask.jsonl",
540
- "source_sha256": "8eccd2067eea7427e6550659653a93a9d883ef531159b8ac806cf2c8fbbdff1c",
541
- "sha256": "8eccd2067eea7427e6550659653a93a9d883ef531159b8ac806cf2c8fbbdff1c",
542
- "source_bytes": 676064,
543
- "bytes": 676064,
544
- "replacements": {},
545
- "rows": 1527
546
- },
547
- {
548
- "path": "coderank/units/gin.jsonl",
549
- "source_sha256": "a5e44ede5d75cbd1f9ae6bf829001717fac12c59741efdb4acffada3d1eb5f8f",
550
- "sha256": "a5e44ede5d75cbd1f9ae6bf829001717fac12c59741efdb4acffada3d1eb5f8f",
551
- "source_bytes": 626917,
552
- "bytes": 626917,
553
- "replacements": {},
554
- "rows": 1550
555
- },
556
- {
557
- "path": "coderank/units/gson.jsonl",
558
- "source_sha256": "6338623a4512d197eb8e2d94da21d139923ff3cb5663ff665c24605fabf26265",
559
- "sha256": "6338623a4512d197eb8e2d94da21d139923ff3cb5663ff665c24605fabf26265",
560
- "source_bytes": 2438824,
561
- "bytes": 2438824,
562
- "replacements": {},
563
- "rows": 3993
564
- },
565
- {
566
- "path": "coderank/units/httpx.jsonl",
567
- "source_sha256": "d8de37959c02c47055e6fc4cb8194ed048fe611811672e3e4bd02a63150531b9",
568
- "sha256": "d8de37959c02c47055e6fc4cb8194ed048fe611811672e3e4bd02a63150531b9",
569
- "source_bytes": 647104,
570
- "bytes": 647104,
571
- "replacements": {},
572
- "rows": 1158
573
- },
574
- {
575
- "path": "coderank/units/okhttp.jsonl",
576
- "source_sha256": "2eb89801e5093e44f49e615e2a3738d20a8f0a89d437227ab2f84c2b1dcc0466",
577
- "sha256": "2eb89801e5093e44f49e615e2a3738d20a8f0a89d437227ab2f84c2b1dcc0466",
578
- "source_bytes": 4687545,
579
- "bytes": 4687545,
580
- "replacements": {},
581
- "rows": 7622
582
- },
583
- {
584
- "path": "coderank/units/requests.jsonl",
585
- "source_sha256": "ea2815d1250b86879a36dc4f62b9f30a87f8ca45159ac3287ef7fbdf70dda28b",
586
- "sha256": "ea2815d1250b86879a36dc4f62b9f30a87f8ca45159ac3287ef7fbdf70dda28b",
587
- "source_bytes": 353227,
588
- "bytes": 353227,
589
- "replacements": {},
590
- "rows": 703
591
- },
592
- {
593
- "path": "coderank/units/ripgrep.jsonl",
594
- "source_sha256": "b22d4bf5fd1a7eb8916b4f44bc71bb3295372b06905b9fbc62c2ff5072cce30d",
595
- "sha256": "b22d4bf5fd1a7eb8916b4f44bc71bb3295372b06905b9fbc62c2ff5072cce30d",
596
- "source_bytes": 1309945,
597
- "bytes": 1309945,
598
- "replacements": {},
599
- "rows": 2802
600
- },
601
- {
602
- "path": "coderank/units/sinatra.jsonl",
603
- "source_sha256": "96a59e2627dc8e546d87dc795f9fd1598297051fc4065991a1cd43becfc52b2d",
604
- "sha256": "96a59e2627dc8e546d87dc795f9fd1598297051fc4065991a1cd43becfc52b2d",
605
- "source_bytes": 474713,
606
- "bytes": 474713,
607
- "replacements": {},
608
- "rows": 987
609
- },
610
- {
611
- "path": "coderank/units/slim.jsonl",
612
- "source_sha256": "8833fcbc679766581cf2e0d2f4afa6067593c420a607f5b5a822181ac54df309",
613
- "sha256": "8833fcbc679766581cf2e0d2f4afa6067593c420a607f5b5a822181ac54df309",
614
- "source_bytes": 631323,
615
- "bytes": 631323,
616
- "replacements": {},
617
- "rows": 884
618
- },
619
- {
620
- "path": "coderank/units/thor.jsonl",
621
- "source_sha256": "c2b13a10ad583904e26a43188e9fad06c508fafe15bf17b8399227a7aadf7a71",
622
- "sha256": "c2b13a10ad583904e26a43188e9fad06c508fafe15bf17b8399227a7aadf7a71",
623
- "source_bytes": 364924,
624
- "bytes": 364924,
625
- "replacements": {},
626
- "rows": 802
627
- },
628
- {
629
- "path": "coderank/units/zod.jsonl",
630
- "source_sha256": "977460c1f255e340e858347ece342efdb0ae458f10a100fdf93c871840dc0045",
631
- "sha256": "977460c1f255e340e858347ece342efdb0ae458f10a100fdf93c871840dc0045",
632
- "source_bytes": 2452362,
633
- "bytes": 2452362,
634
- "replacements": {},
635
- "rows": 8253
636
- },
637
- {
638
- "path": "coderank/dumps/project_search-docs-axios.jsonl",
639
- "source_sha256": "5bcfcc5cf524bf153544e076f8ee04377dd328a087af6655a9ac86c95ec29559",
640
- "sha256": "5bcfcc5cf524bf153544e076f8ee04377dd328a087af6655a9ac86c95ec29559",
641
- "source_bytes": 4825,
642
- "bytes": 4825,
643
- "replacements": {},
644
- "rows": 5
645
- },
646
- {
647
- "path": "coderank/dumps/project_search-docs-axum.jsonl",
648
- "source_sha256": "eeb63cd18accad559cb170cb32c60909acbea46105ed77d3a5c0db74ca190f3a",
649
- "sha256": "eeb63cd18accad559cb170cb32c60909acbea46105ed77d3a5c0db74ca190f3a",
650
- "source_bytes": 23288,
651
- "bytes": 23288,
652
- "replacements": {},
653
- "rows": 28
654
- },
655
- {
656
- "path": "coderank/dumps/project_search-docs-cobra.jsonl",
657
- "source_sha256": "883f73dc55ac6fd11419cda77b6ff87a7ca6c58fdb0dcb6956b4e004cf44098e",
658
- "sha256": "883f73dc55ac6fd11419cda77b6ff87a7ca6c58fdb0dcb6956b4e004cf44098e",
659
- "source_bytes": 9382,
660
- "bytes": 9382,
661
- "replacements": {},
662
- "rows": 15
663
- },
664
- {
665
- "path": "coderank/dumps/project_search-docs-fd.jsonl",
666
- "source_sha256": "79b5695acd315bb10c178c51d7c717014d5fcb3a9daeaad780bde438a646e8ee",
667
- "sha256": "79b5695acd315bb10c178c51d7c717014d5fcb3a9daeaad780bde438a646e8ee",
668
- "source_bytes": 6762,
669
- "bytes": 6762,
670
- "replacements": {},
671
- "rows": 8
672
- },
673
- {
674
- "path": "coderank/dumps/project_search-docs-gin.jsonl",
675
- "source_sha256": "c605944583ec4cb997a63d1db919d19d51958deec276325aa798f737d0aee35c",
676
- "sha256": "c605944583ec4cb997a63d1db919d19d51958deec276325aa798f737d0aee35c",
677
- "source_bytes": 108622,
678
- "bytes": 108622,
679
- "replacements": {},
680
- "rows": 41
681
- },
682
- {
683
- "path": "coderank/dumps/project_search-docs-gson.jsonl",
684
- "source_sha256": "810915dab0f649e75d437b4c2399ca9b46ca71884d23f926c43ef205a5c53d57",
685
- "sha256": "810915dab0f649e75d437b4c2399ca9b46ca71884d23f926c43ef205a5c53d57",
686
- "source_bytes": 67318,
687
- "bytes": 67318,
688
- "replacements": {},
689
- "rows": 44
690
- },
691
- {
692
- "path": "coderank/dumps/project_search-docs-httpx.jsonl",
693
- "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
694
- "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
695
- "source_bytes": 0,
696
- "bytes": 0,
697
- "replacements": {},
698
- "rows": 0
699
- },
700
- {
701
- "path": "coderank/dumps/project_search-docs-okhttp.jsonl",
702
- "source_sha256": "509e68398d0e8b22b3b60232132e84f671f2ddcc556a41aa167dc2888f274c72",
703
- "sha256": "509e68398d0e8b22b3b60232132e84f671f2ddcc556a41aa167dc2888f274c72",
704
- "source_bytes": 134009,
705
- "bytes": 134009,
706
- "replacements": {},
707
- "rows": 77
708
- },
709
- {
710
- "path": "coderank/dumps/project_search-docs-ripgrep.jsonl",
711
- "source_sha256": "3375be55202398c3bac09c742659540fad16c2905def71335990b3d146b970ab",
712
- "sha256": "3375be55202398c3bac09c742659540fad16c2905def71335990b3d146b970ab",
713
- "source_bytes": 355627,
714
- "bytes": 355627,
715
- "replacements": {},
716
- "rows": 150
717
- },
718
- {
719
- "path": "coderank/dumps/project_search-docs-sinatra.jsonl",
720
- "source_sha256": "411077eb35f38a817a11b4373209967a1cd6364149b89d55d43806bdd476e1cf",
721
- "sha256": "411077eb35f38a817a11b4373209967a1cd6364149b89d55d43806bdd476e1cf",
722
- "source_bytes": 8336,
723
- "bytes": 8336,
724
- "replacements": {},
725
- "rows": 8
726
- },
727
- {
728
- "path": "coderank/dumps/project_search-docs-slim.jsonl",
729
- "source_sha256": "4d83ae26b127a67ba643260ba431a3d1ca0f3b84261fe2f7cedb92ada74c7312",
730
- "sha256": "4d83ae26b127a67ba643260ba431a3d1ca0f3b84261fe2f7cedb92ada74c7312",
731
- "source_bytes": 4638,
732
- "bytes": 4638,
733
- "replacements": {},
734
- "rows": 5
735
- },
736
- {
737
- "path": "coderank/dumps/project_search-docs-thor.jsonl",
738
- "source_sha256": "7e5bf295be859366377883ffdced6715f952b1f743f95eb10d7bf66fc278daac",
739
- "sha256": "7e5bf295be859366377883ffdced6715f952b1f743f95eb10d7bf66fc278daac",
740
- "source_bytes": 6644,
741
- "bytes": 6644,
742
- "replacements": {},
743
- "rows": 8
744
- },
745
- {
746
- "path": "coderank/dumps/project_search-docs-zod.jsonl",
747
- "source_sha256": "cf7b8e1cfc6db1be72ca542c1073e556121046b92c93a1b5646f1b7816060a64",
748
- "sha256": "cf7b8e1cfc6db1be72ca542c1073e556121046b92c93a1b5646f1b7816060a64",
749
- "source_bytes": 107017,
750
- "bytes": 107017,
751
- "replacements": {},
752
- "rows": 74
753
- },
754
- {
755
- "path": "coderank/dumps/project_search-model-axios.jsonl",
756
- "source_sha256": "e8665c1b219ba19ed8b9efc04ec9a59abbd5b3e5ac7245244ab46e0907fb3773",
757
- "sha256": "e8665c1b219ba19ed8b9efc04ec9a59abbd5b3e5ac7245244ab46e0907fb3773",
758
- "source_bytes": 1102506,
759
- "bytes": 1102506,
760
- "replacements": {},
761
- "rows": 291
762
- },
763
- {
764
- "path": "coderank/dumps/project_search-model-axum.jsonl",
765
- "source_sha256": "e396c406a99d6b0b2adc663b07b6902aa1dba2738c0990fb6ee5cba8c7eb8b71",
766
- "sha256": "e396c406a99d6b0b2adc663b07b6902aa1dba2738c0990fb6ee5cba8c7eb8b71",
767
- "source_bytes": 970171,
768
- "bytes": 970171,
769
- "replacements": {},
770
- "rows": 332
771
- },
772
- {
773
- "path": "coderank/dumps/project_search-model-cobra.jsonl",
774
- "source_sha256": "f758dd450ffde967901524f830bb82809b23b74676d7dfd8192ef90f2525a109",
775
- "sha256": "f758dd450ffde967901524f830bb82809b23b74676d7dfd8192ef90f2525a109",
776
- "source_bytes": 345239,
777
- "bytes": 345239,
778
- "replacements": {},
779
- "rows": 140
780
- },
781
- {
782
- "path": "coderank/dumps/project_search-model-express.jsonl",
783
- "source_sha256": "4e10be53a860d918c974c6f7c0df495b2057ba333ba32f6d5f27a7be9eb6d07b",
784
- "sha256": "4e10be53a860d918c974c6f7c0df495b2057ba333ba32f6d5f27a7be9eb6d07b",
785
- "source_bytes": 48491,
786
- "bytes": 48491,
787
- "replacements": {},
788
- "rows": 31
789
- },
790
- {
791
- "path": "coderank/dumps/project_search-model-fd.jsonl",
792
- "source_sha256": "8e49366dfbdb60e101178ad8176dcc457debd759037881dec2b34c8cf0ff76d1",
793
- "sha256": "8e49366dfbdb60e101178ad8176dcc457debd759037881dec2b34c8cf0ff76d1",
794
- "source_bytes": 121235,
795
- "bytes": 121235,
796
- "replacements": {},
797
- "rows": 68
798
- },
799
- {
800
- "path": "coderank/dumps/project_search-model-flask.jsonl",
801
- "source_sha256": "4f296ad85a8098aa8157d1c961594a37a2e161e0243532d33a44697d144a44bd",
802
- "sha256": "4f296ad85a8098aa8157d1c961594a37a2e161e0243532d33a44697d144a44bd",
803
- "source_bytes": 710235,
804
- "bytes": 710235,
805
- "replacements": {},
806
- "rows": 320
807
- },
808
- {
809
- "path": "coderank/dumps/project_search-model-gin.jsonl",
810
- "source_sha256": "d2dce8f7c3ebe84d49bc0be4bacc963c91534c8e667ac465c195c329e25ef447",
811
- "sha256": "d2dce8f7c3ebe84d49bc0be4bacc963c91534c8e667ac465c195c329e25ef447",
812
- "source_bytes": 596467,
813
- "bytes": 596467,
814
- "replacements": {},
815
- "rows": 325
816
- },
817
- {
818
- "path": "coderank/dumps/project_search-model-gson.jsonl",
819
- "source_sha256": "624c9436208e17e12e5455640e9247a6dd352746093b54baf1a297bc33766893",
820
- "sha256": "624c9436208e17e12e5455640e9247a6dd352746093b54baf1a297bc33766893",
821
- "source_bytes": 918984,
822
- "bytes": 918984,
823
- "replacements": {},
824
- "rows": 275
825
- },
826
- {
827
- "path": "coderank/dumps/project_search-model-httpx.jsonl",
828
- "source_sha256": "5f3bc5225a03041c130090e21a92cf4178410aed4e1d13f04c7c3b3d5a1cc597",
829
- "sha256": "5f3bc5225a03041c130090e21a92cf4178410aed4e1d13f04c7c3b3d5a1cc597",
830
- "source_bytes": 442096,
831
- "bytes": 442096,
832
- "replacements": {},
833
- "rows": 222
834
- },
835
- {
836
- "path": "coderank/dumps/project_search-model-okhttp.jsonl",
837
- "source_sha256": "ab119fcc6ec66db1962cc362fad8657ee7c78fc409c1de8923b36fb8261e58b6",
838
- "sha256": "ab119fcc6ec66db1962cc362fad8657ee7c78fc409c1de8923b36fb8261e58b6",
839
- "source_bytes": 937036,
840
- "bytes": 937036,
841
- "replacements": {},
842
- "rows": 254
843
- },
844
- {
845
- "path": "coderank/dumps/project_search-model-requests.jsonl",
846
- "source_sha256": "150474fd41d416e62c33c2db20573117d4b2dc1e4ee0c06646e396554f83984e",
847
- "sha256": "150474fd41d416e62c33c2db20573117d4b2dc1e4ee0c06646e396554f83984e",
848
- "source_bytes": 253551,
849
- "bytes": 253551,
850
- "replacements": {},
851
- "rows": 151
852
- },
853
- {
854
- "path": "coderank/dumps/project_search-model-ripgrep.jsonl",
855
- "source_sha256": "c099cfc6a427392781529bd41485d80d54b0ecd5b73eb1a32877f629f7f25b65",
856
- "sha256": "c099cfc6a427392781529bd41485d80d54b0ecd5b73eb1a32877f629f7f25b65",
857
- "source_bytes": 1119400,
858
- "bytes": 1119400,
859
- "replacements": {},
860
- "rows": 364
861
- },
862
- {
863
- "path": "coderank/dumps/project_search-model-sinatra.jsonl",
864
- "source_sha256": "b76fb56eba15cb615fc7ae548f91928dc3be4df71d051248c48171529cb0b699",
865
- "sha256": "b76fb56eba15cb615fc7ae548f91928dc3be4df71d051248c48171529cb0b699",
866
- "source_bytes": 293598,
867
- "bytes": 293598,
868
- "replacements": {},
869
- "rows": 152
870
- },
871
- {
872
- "path": "coderank/dumps/project_search-model-slim.jsonl",
873
- "source_sha256": "a7836761487527798a4d8298cfcc7eb2582760658a6e09d2365cc56c230fe87e",
874
- "sha256": "a7836761487527798a4d8298cfcc7eb2582760658a6e09d2365cc56c230fe87e",
875
- "source_bytes": 723386,
876
- "bytes": 723386,
877
- "replacements": {},
878
- "rows": 225
879
- },
880
- {
881
- "path": "coderank/dumps/project_search-model-thor.jsonl",
882
- "source_sha256": "0b84460ad6c4e8cac36686add70716b975cdb1c6c2f8059d818eb8e28afb2222",
883
- "sha256": "0b84460ad6c4e8cac36686add70716b975cdb1c6c2f8059d818eb8e28afb2222",
884
- "source_bytes": 406744,
885
- "bytes": 406744,
886
- "replacements": {},
887
- "rows": 173
888
- },
889
- {
890
- "path": "coderank/dumps/project_search-model-zod.jsonl",
891
- "source_sha256": "580632b2fa373fd9960ddcd85c42aa8aaa6d2c3e76d9b6219af67098b4e15e2e",
892
- "sha256": "580632b2fa373fd9960ddcd85c42aa8aaa6d2c3e76d9b6219af67098b4e15e2e",
893
- "source_bytes": 1964854,
894
- "bytes": 1964854,
895
- "replacements": {},
896
- "rows": 464
897
- },
898
- {
899
- "path": "coderank/dumps/repomap_tags-docs-axios.jsonl",
900
- "source_sha256": "f695f184413b1cbcae84bff755f39d46873cae49064f6fa3ed17f1cfa2c9a9b4",
901
- "sha256": "f695f184413b1cbcae84bff755f39d46873cae49064f6fa3ed17f1cfa2c9a9b4",
902
- "source_bytes": 8837,
903
- "bytes": 8837,
904
- "replacements": {},
905
- "rows": 2
906
- },
907
- {
908
- "path": "coderank/dumps/repomap_tags-docs-axum.jsonl",
909
- "source_sha256": "3ceb783a0aa1379cd57278686978e013e61beb7d284e3bfc0512120366c2883e",
910
- "sha256": "3ceb783a0aa1379cd57278686978e013e61beb7d284e3bfc0512120366c2883e",
911
- "source_bytes": 656279,
912
- "bytes": 656279,
913
- "replacements": {},
914
- "rows": 37
915
- },
916
- {
917
- "path": "coderank/dumps/repomap_tags-docs-cobra.jsonl",
918
- "source_sha256": "57149dc1d9a137a8f571e06e11e06f5df8a9aade5faf3ca6ade8a98733529807",
919
- "sha256": "57149dc1d9a137a8f571e06e11e06f5df8a9aade5faf3ca6ade8a98733529807",
920
- "source_bytes": 19641,
921
- "bytes": 19641,
922
- "replacements": {},
923
- "rows": 1
924
- },
925
- {
926
- "path": "coderank/dumps/repomap_tags-docs-fd.jsonl",
927
- "source_sha256": "d27ff9f47ef9045f82116151ce938dcca0da3b0329c74bafa34c7ed1ed0310aa",
928
- "sha256": "d27ff9f47ef9045f82116151ce938dcca0da3b0329c74bafa34c7ed1ed0310aa",
929
- "source_bytes": 26416,
930
- "bytes": 26416,
931
- "replacements": {},
932
- "rows": 4
933
- },
934
- {
935
- "path": "coderank/dumps/repomap_tags-docs-gin.jsonl",
936
- "source_sha256": "d58baa5d84cd9443c199ef40d01b46dda3e63f9cc225ccdd2d0075e96777a9f8",
937
- "sha256": "d58baa5d84cd9443c199ef40d01b46dda3e63f9cc225ccdd2d0075e96777a9f8",
938
- "source_bytes": 773653,
939
- "bytes": 773653,
940
- "replacements": {},
941
- "rows": 23
942
- },
943
- {
944
- "path": "coderank/dumps/repomap_tags-docs-gson.jsonl",
945
- "source_sha256": "489c03a4cf47fab381c57c60afb15c773e704f8794e97d128c826d0b4b6ce059",
946
- "sha256": "489c03a4cf47fab381c57c60afb15c773e704f8794e97d128c826d0b4b6ce059",
947
- "source_bytes": 356554,
948
- "bytes": 356554,
949
- "replacements": {},
950
- "rows": 11
951
- },
952
- {
953
- "path": "coderank/dumps/repomap_tags-docs-httpx.jsonl",
954
- "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
955
- "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
956
- "source_bytes": 0,
957
- "bytes": 0,
958
- "replacements": {},
959
- "rows": 0
960
- },
961
- {
962
- "path": "coderank/dumps/repomap_tags-docs-okhttp.jsonl",
963
- "source_sha256": "6f60069dbcfd26a2e63bfbed00f18609926c3178a80600a0dc05bb7d737688b7",
964
- "sha256": "6f60069dbcfd26a2e63bfbed00f18609926c3178a80600a0dc05bb7d737688b7",
965
- "source_bytes": 4721408,
966
- "bytes": 4721408,
967
- "replacements": {},
968
- "rows": 69
969
- },
970
- {
971
- "path": "coderank/dumps/repomap_tags-docs-ripgrep.jsonl",
972
- "source_sha256": "70e1458de74c9c0a31d0351de01cdab23d435ae28b9ac383d68459c58458e30b",
973
- "sha256": "70e1458de74c9c0a31d0351de01cdab23d435ae28b9ac383d68459c58458e30b",
974
- "source_bytes": 10140158,
975
- "bytes": 10140158,
976
- "replacements": {},
977
- "rows": 117
978
- },
979
- {
980
- "path": "coderank/dumps/repomap_tags-docs-sinatra.jsonl",
981
- "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
982
- "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
983
- "source_bytes": 0,
984
- "bytes": 0,
985
- "replacements": {},
986
- "rows": 0
987
- },
988
- {
989
- "path": "coderank/dumps/repomap_tags-docs-slim.jsonl",
990
- "source_sha256": "e4ee8169ee1d5b267e5a3243551a5806a787e156470418940fbbb88c17200f48",
991
- "sha256": "e4ee8169ee1d5b267e5a3243551a5806a787e156470418940fbbb88c17200f48",
992
- "source_bytes": 366898,
993
- "bytes": 366898,
994
- "replacements": {},
995
- "rows": 10
996
- },
997
- {
998
- "path": "coderank/dumps/repomap_tags-docs-thor.jsonl",
999
- "source_sha256": "673053bab8ebff64080b7a45a86827da139f245f230cd11814ec8a6437d3f724",
1000
- "sha256": "673053bab8ebff64080b7a45a86827da139f245f230cd11814ec8a6437d3f724",
1001
- "source_bytes": 205968,
1002
- "bytes": 205968,
1003
- "replacements": {},
1004
- "rows": 11
1005
- },
1006
- {
1007
- "path": "coderank/dumps/repomap_tags-docs-zod.jsonl",
1008
- "source_sha256": "259bd08ef58038d41f36e3d2c64110417e9891d761ef8375121f0e3541e14eea",
1009
- "sha256": "259bd08ef58038d41f36e3d2c64110417e9891d761ef8375121f0e3541e14eea",
1010
- "source_bytes": 1865916,
1011
- "bytes": 1865916,
1012
- "replacements": {},
1013
- "rows": 25
1014
- },
1015
- {
1016
- "path": "coderank/dumps/repomap_tags-model-axios.jsonl",
1017
- "source_sha256": "229737da35f6ebee09806e031df8d5ac2030208c0786e148d2e7129ea1fff941",
1018
- "sha256": "229737da35f6ebee09806e031df8d5ac2030208c0786e148d2e7129ea1fff941",
1019
- "source_bytes": 870239,
1020
- "bytes": 870239,
1021
- "replacements": {},
1022
- "rows": 56
1023
- },
1024
- {
1025
- "path": "coderank/dumps/repomap_tags-model-axum.jsonl",
1026
- "source_sha256": "7d4a8f7d2e23d2d5b9d62713c290ad1f02a59c04d87bb5fe1b8008d8fd89260a",
1027
- "sha256": "7d4a8f7d2e23d2d5b9d62713c290ad1f02a59c04d87bb5fe1b8008d8fd89260a",
1028
- "source_bytes": 12831850,
1029
- "bytes": 12831850,
1030
- "replacements": {},
1031
- "rows": 526
1032
- },
1033
- {
1034
- "path": "coderank/dumps/repomap_tags-model-cobra.jsonl",
1035
- "source_sha256": "b85e1bf8c68a9ff0a069fc35d8b901172f4ce9d6cdd90ea70e6a0d02072bbb9a",
1036
- "sha256": "b85e1bf8c68a9ff0a069fc35d8b901172f4ce9d6cdd90ea70e6a0d02072bbb9a",
1037
- "source_bytes": 576429,
1038
- "bytes": 576429,
1039
- "replacements": {},
1040
- "rows": 30
1041
- },
1042
- {
1043
- "path": "coderank/dumps/repomap_tags-model-express.jsonl",
1044
- "source_sha256": "288ac59738e652f50d3d3f2c640fd51f649454f1c4341af7ba1c8bab91c66308",
1045
- "sha256": "288ac59738e652f50d3d3f2c640fd51f649454f1c4341af7ba1c8bab91c66308",
1046
- "source_bytes": 68684,
1047
- "bytes": 68684,
1048
- "replacements": {},
1049
- "rows": 20
1050
- },
1051
- {
1052
- "path": "coderank/dumps/repomap_tags-model-fd.jsonl",
1053
- "source_sha256": "679ac2ecfe940668709193003782e887b8278484d18a93b939203603af4caa37",
1054
- "sha256": "679ac2ecfe940668709193003782e887b8278484d18a93b939203603af4caa37",
1055
- "source_bytes": 1246280,
1056
- "bytes": 1246280,
1057
- "replacements": {},
1058
- "rows": 78
1059
- },
1060
- {
1061
- "path": "coderank/dumps/repomap_tags-model-flask.jsonl",
1062
- "source_sha256": "d448ef3155927857cfd90ef32be430512e41c2dab61e1c33bedbcfabfacc28e1",
1063
- "sha256": "d448ef3155927857cfd90ef32be430512e41c2dab61e1c33bedbcfabfacc28e1",
1064
- "source_bytes": 3001097,
1065
- "bytes": 3001097,
1066
- "replacements": {},
1067
- "rows": 188
1068
- },
1069
- {
1070
- "path": "coderank/dumps/repomap_tags-model-gin.jsonl",
1071
- "source_sha256": "e5b349569d88f7e2d29cfc33035b1b50c9df529fc10e2bca502b4d0cf6691b92",
1072
- "sha256": "e5b349569d88f7e2d29cfc33035b1b50c9df529fc10e2bca502b4d0cf6691b92",
1073
- "source_bytes": 23956210,
1074
- "bytes": 23956210,
1075
- "replacements": {},
1076
- "rows": 279
1077
- },
1078
- {
1079
- "path": "coderank/dumps/repomap_tags-model-gson.jsonl",
1080
- "source_sha256": "53556264e7ee14cc7a84e5c4dfb736969e89e83c9440e69959ce201d99dcc3a6",
1081
- "sha256": "53556264e7ee14cc7a84e5c4dfb736969e89e83c9440e69959ce201d99dcc3a6",
1082
- "source_bytes": 5640978,
1083
- "bytes": 5640978,
1084
- "replacements": {},
1085
- "rows": 158
1086
- },
1087
- {
1088
- "path": "coderank/dumps/repomap_tags-model-httpx.jsonl",
1089
- "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
1090
- "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
1091
- "source_bytes": 0,
1092
- "bytes": 0,
1093
- "replacements": {},
1094
- "rows": 0
1095
- },
1096
- {
1097
- "path": "coderank/dumps/repomap_tags-model-okhttp.jsonl",
1098
- "source_sha256": "042607519a6f51d10b301c46307d562708461b2c06c1aeda63dc63004ac3f5ba",
1099
- "sha256": "042607519a6f51d10b301c46307d562708461b2c06c1aeda63dc63004ac3f5ba",
1100
- "source_bytes": 16287219,
1101
- "bytes": 16287219,
1102
- "replacements": {},
1103
- "rows": 266
1104
- },
1105
- {
1106
- "path": "coderank/dumps/repomap_tags-model-requests.jsonl",
1107
- "source_sha256": "a8c8f83ff7593156a95a317adc50375b080ab18a30321b33722b59a879d3545d",
1108
- "sha256": "a8c8f83ff7593156a95a317adc50375b080ab18a30321b33722b59a879d3545d",
1109
- "source_bytes": 88191,
1110
- "bytes": 88191,
1111
- "replacements": {},
1112
- "rows": 24
1113
- },
1114
- {
1115
- "path": "coderank/dumps/repomap_tags-model-ripgrep.jsonl",
1116
- "source_sha256": "d4e26d6f5eb13c5b3e65306c74f2f413e7f539af54fde4a5e564785c766b0de9",
1117
- "sha256": "d4e26d6f5eb13c5b3e65306c74f2f413e7f539af54fde4a5e564785c766b0de9",
1118
- "source_bytes": 25736420,
1119
- "bytes": 25736420,
1120
- "replacements": {},
1121
- "rows": 344
1122
- },
1123
- {
1124
- "path": "coderank/dumps/repomap_tags-model-sinatra.jsonl",
1125
- "source_sha256": "6642d6f9d816bd9ecb343de7b5d1265237286a79292e014fdb75da93e54397c2",
1126
- "sha256": "6642d6f9d816bd9ecb343de7b5d1265237286a79292e014fdb75da93e54397c2",
1127
- "source_bytes": 2557204,
1128
- "bytes": 2557204,
1129
- "replacements": {},
1130
- "rows": 74
1131
- },
1132
- {
1133
- "path": "coderank/dumps/repomap_tags-model-slim.jsonl",
1134
- "source_sha256": "2cbe2dd858509e366c346b743c607a6cb34a591e26f6488c8a4952e798dda0ee",
1135
- "sha256": "2cbe2dd858509e366c346b743c607a6cb34a591e26f6488c8a4952e798dda0ee",
1136
- "source_bytes": 9574837,
1137
- "bytes": 9574837,
1138
- "replacements": {},
1139
- "rows": 333
1140
- },
1141
- {
1142
- "path": "coderank/dumps/repomap_tags-model-thor.jsonl",
1143
- "source_sha256": "aa766f7fab0130940bcdd273653b64caebbf71456c7823931a7ca7757c21c528",
1144
- "sha256": "aa766f7fab0130940bcdd273653b64caebbf71456c7823931a7ca7757c21c528",
1145
- "source_bytes": 3223571,
1146
- "bytes": 3223571,
1147
- "replacements": {},
1148
- "rows": 204
1149
- },
1150
- {
1151
- "path": "coderank/dumps/repomap_tags-model-zod.jsonl",
1152
- "source_sha256": "1e1b341d493aa6cba1951b89570320ce59649192d3d1985f581adf2bd0274533",
1153
- "sha256": "1e1b341d493aa6cba1951b89570320ce59649192d3d1985f581adf2bd0274533",
1154
- "source_bytes": 9490746,
1155
- "bytes": 9490746,
1156
- "replacements": {},
1157
- "rows": 129
1158
- },
1159
- {
1160
- "path": "coderank/dumps/summarize_definitions-docs-axios.jsonl",
1161
- "source_sha256": "d4de787e189714deabdedf947cf35132a05bbd2d2c94f7e80e46a3e42718033a",
1162
- "sha256": "d4de787e189714deabdedf947cf35132a05bbd2d2c94f7e80e46a3e42718033a",
1163
- "source_bytes": 217564,
1164
- "bytes": 217564,
1165
- "replacements": {},
1166
- "rows": 9
1167
- },
1168
- {
1169
- "path": "coderank/dumps/summarize_definitions-docs-axum.jsonl",
1170
- "source_sha256": "d134a1ba87d1e2853ad108fad077c953506fba00495cb7fe34bbdf3931bb6693",
1171
- "sha256": "d134a1ba87d1e2853ad108fad077c953506fba00495cb7fe34bbdf3931bb6693",
1172
- "source_bytes": 3713826,
1173
- "bytes": 3713826,
1174
- "replacements": {},
1175
- "rows": 53
1176
- },
1177
- {
1178
- "path": "coderank/dumps/summarize_definitions-docs-cobra.jsonl",
1179
- "source_sha256": "817c853ccbe61f4e3da7c0aeab6ff103d18f8eaa6fbea67e9ebeefbc3857bf6e",
1180
- "sha256": "817c853ccbe61f4e3da7c0aeab6ff103d18f8eaa6fbea67e9ebeefbc3857bf6e",
1181
- "source_bytes": 4646818,
1182
- "bytes": 4646818,
1183
- "replacements": {},
1184
- "rows": 33
1185
- },
1186
- {
1187
- "path": "coderank/dumps/summarize_definitions-docs-fd.jsonl",
1188
- "source_sha256": "88a4302822ceb41ca6fc1e7bc2743e0ae25fe34b180df36d6377cb884ba0a416",
1189
- "sha256": "88a4302822ceb41ca6fc1e7bc2743e0ae25fe34b180df36d6377cb884ba0a416",
1190
- "source_bytes": 737524,
1191
- "bytes": 737524,
1192
- "replacements": {},
1193
- "rows": 15
1194
- },
1195
- {
1196
- "path": "coderank/dumps/summarize_definitions-docs-gin.jsonl",
1197
- "source_sha256": "993c85a1c454add4f73beb087bf8ca983dd78fd648c7b58374a0ea472ad0df08",
1198
- "sha256": "993c85a1c454add4f73beb087bf8ca983dd78fd648c7b58374a0ea472ad0df08",
1199
- "source_bytes": 13528952,
1200
- "bytes": 13528952,
1201
- "replacements": {},
1202
- "rows": 72
1203
- },
1204
- {
1205
- "path": "coderank/dumps/summarize_definitions-docs-gson.jsonl",
1206
- "source_sha256": "a36dac11920e5af69173534bfed697ffb3747f712ec34569faee02fb576236e7",
1207
- "sha256": "a36dac11920e5af69173534bfed697ffb3747f712ec34569faee02fb576236e7",
1208
- "source_bytes": 19784143,
1209
- "bytes": 19784143,
1210
- "replacements": {},
1211
- "rows": 73
1212
- },
1213
- {
1214
- "path": "coderank/dumps/summarize_definitions-docs-httpx.jsonl",
1215
- "source_sha256": "fd9307f262b8c682e53397c48d89b588b910771ef0a57c378fe15b4ab203e43c",
1216
- "sha256": "fd9307f262b8c682e53397c48d89b588b910771ef0a57c378fe15b4ab203e43c",
1217
- "source_bytes": 91601,
1218
- "bytes": 91601,
1219
- "replacements": {},
1220
- "rows": 1
1221
- },
1222
- {
1223
- "path": "coderank/dumps/summarize_definitions-docs-okhttp.jsonl",
1224
- "source_sha256": "755045282839ce07e85b1daaf542c599aaa722dec6a2bdfa125b124a9b755d56",
1225
- "sha256": "755045282839ce07e85b1daaf542c599aaa722dec6a2bdfa125b124a9b755d56",
1226
- "source_bytes": 34340070,
1227
- "bytes": 34340070,
1228
- "replacements": {},
1229
- "rows": 184
1230
- },
1231
- {
1232
- "path": "coderank/dumps/summarize_definitions-docs-ripgrep.jsonl",
1233
- "source_sha256": "80c904159630f2925c9defcfb9fe81c0f596526a12c815d8fb282310c7436910",
1234
- "sha256": "80c904159630f2925c9defcfb9fe81c0f596526a12c815d8fb282310c7436910",
1235
- "source_bytes": 18198877,
1236
- "bytes": 18198877,
1237
- "replacements": {},
1238
- "rows": 243
1239
- },
1240
- {
1241
- "path": "coderank/dumps/summarize_definitions-docs-sinatra.jsonl",
1242
- "source_sha256": "6352430b75ce4874c8d62cad2d406e84f342d4e2b483685a22fa9c536737fa99",
1243
- "sha256": "6352430b75ce4874c8d62cad2d406e84f342d4e2b483685a22fa9c536737fa99",
1244
- "source_bytes": 1859089,
1245
- "bytes": 1859089,
1246
- "replacements": {},
1247
- "rows": 29
1248
- },
1249
- {
1250
- "path": "coderank/dumps/summarize_definitions-docs-slim.jsonl",
1251
- "source_sha256": "412176f1689a91e4ebc624ae72a09adf2f8c0e58a0fa4a7a14abbda19f0bc7e4",
1252
- "sha256": "412176f1689a91e4ebc624ae72a09adf2f8c0e58a0fa4a7a14abbda19f0bc7e4",
1253
- "source_bytes": 367106,
1254
- "bytes": 367106,
1255
- "replacements": {},
1256
- "rows": 10
1257
- },
1258
- {
1259
- "path": "coderank/dumps/summarize_definitions-docs-thor.jsonl",
1260
- "source_sha256": "f87c6e2767f4f2a6abc3ce6698a9464c6e31590b0ac6d6a68243d18af49e0906",
1261
- "sha256": "f87c6e2767f4f2a6abc3ce6698a9464c6e31590b0ac6d6a68243d18af49e0906",
1262
- "source_bytes": 1843809,
1263
- "bytes": 1843809,
1264
- "replacements": {},
1265
- "rows": 24
1266
- },
1267
- {
1268
- "path": "coderank/dumps/summarize_definitions-docs-zod.jsonl",
1269
- "source_sha256": "eb1f452a1689778938fd7bf6c638ea3ab9cd4229f41b5233169b55a67a209dcf",
1270
- "sha256": "eb1f452a1689778938fd7bf6c638ea3ab9cd4229f41b5233169b55a67a209dcf",
1271
- "source_bytes": 74192757,
1272
- "bytes": 74192757,
1273
- "replacements": {},
1274
- "rows": 187
1275
- },
1276
- {
1277
- "path": "coderank/dumps/summarize_definitions-model-axios.jsonl",
1278
- "source_sha256": "eb73ad39e04f98f3b7e63e4e257bd3eb74188fca2cc72429c9d73c67f3906d59",
1279
- "sha256": "eb73ad39e04f98f3b7e63e4e257bd3eb74188fca2cc72429c9d73c67f3906d59",
1280
- "source_bytes": 20888356,
1281
- "bytes": 20888356,
1282
- "replacements": {},
1283
- "rows": 583
1284
- },
1285
- {
1286
- "path": "coderank/dumps/summarize_definitions-model-axum.jsonl",
1287
- "source_sha256": "504fb5db950d152ea824e0f1362d96bc77d444e21b666133b304d6f2b6b2ecb4",
1288
- "sha256": "504fb5db950d152ea824e0f1362d96bc77d444e21b666133b304d6f2b6b2ecb4",
1289
- "source_bytes": 34071961,
1290
- "bytes": 34071961,
1291
- "replacements": {},
1292
- "rows": 674
1293
- },
1294
- {
1295
- "path": "coderank/dumps/summarize_definitions-model-cobra.jsonl",
1296
- "source_sha256": "44dbd3241d5c23764f8382568a90a195d36b8d857b7433be154e07fabbe43121",
1297
- "sha256": "44dbd3241d5c23764f8382568a90a195d36b8d857b7433be154e07fabbe43121",
1298
- "source_bytes": 37147184,
1299
- "bytes": 37147184,
1300
- "replacements": {},
1301
- "rows": 287
1302
- },
1303
- {
1304
- "path": "coderank/dumps/summarize_definitions-model-express.jsonl",
1305
- "source_sha256": "be55a20fc30136536ac8b409bed5a2d6422b5d6f6c9b81125e18acbc3aced45f",
1306
- "sha256": "be55a20fc30136536ac8b409bed5a2d6422b5d6f6c9b81125e18acbc3aced45f",
1307
- "source_bytes": 1345642,
1308
- "bytes": 1345642,
1309
- "replacements": {},
1310
- "rows": 98
1311
- },
1312
- {
1313
- "path": "coderank/dumps/summarize_definitions-model-fd.jsonl",
1314
- "source_sha256": "4a432ad7c34be7a1b759935ce218c32ef2615231d0f851ba3d752f8c5f7e84ec",
1315
- "sha256": "4a432ad7c34be7a1b759935ce218c32ef2615231d0f851ba3d752f8c5f7e84ec",
1316
- "source_bytes": 8410402,
1317
- "bytes": 8410402,
1318
- "replacements": {},
1319
- "rows": 214
1320
- },
1321
- {
1322
- "path": "coderank/dumps/summarize_definitions-model-flask.jsonl",
1323
- "source_sha256": "8eb0901688c86c8f0c595e02c57eb6b63973ab5473bccf3804fe4191f584cdf3",
1324
- "sha256": "8eb0901688c86c8f0c595e02c57eb6b63973ab5473bccf3804fe4191f584cdf3",
1325
- "source_bytes": 71779513,
1326
- "bytes": 71779513,
1327
- "replacements": {},
1328
- "rows": 779
1329
- },
1330
- {
1331
- "path": "coderank/dumps/summarize_definitions-model-gin.jsonl",
1332
- "source_sha256": "3678538d2e621e5f04beb8cb764f9931583a03d141f5b4d98dcb2a64b5e95c97",
1333
- "sha256": "3678538d2e621e5f04beb8cb764f9931583a03d141f5b4d98dcb2a64b5e95c97",
1334
- "source_bytes": 126335915,
1335
- "bytes": 126335915,
1336
- "replacements": {},
1337
- "rows": 702
1338
- },
1339
- {
1340
- "path": "coderank/dumps/summarize_definitions-model-gson.jsonl",
1341
- "source_sha256": "2f41f8212572e84bfe3e6547fac6862e3ca8563ebae90c69b322d95c069939a6",
1342
- "sha256": "2f41f8212572e84bfe3e6547fac6862e3ca8563ebae90c69b322d95c069939a6",
1343
- "source_bytes": 166144411,
1344
- "bytes": 166144411,
1345
- "replacements": {},
1346
- "rows": 584
1347
- },
1348
- {
1349
- "path": "coderank/dumps/summarize_definitions-model-httpx.jsonl",
1350
- "source_sha256": "5427da36c7f10eb4aeffa4b22b2dc56fbc8f022b753e519b94b359941f4da9cc",
1351
- "sha256": "5427da36c7f10eb4aeffa4b22b2dc56fbc8f022b753e519b94b359941f4da9cc",
1352
- "source_bytes": 47838081,
1353
- "bytes": 47838081,
1354
- "replacements": {},
1355
- "rows": 547
1356
- },
1357
- {
1358
- "path": "coderank/dumps/summarize_definitions-model-okhttp.jsonl",
1359
- "source_sha256": "9e1c2de08296f90b4f61849b10330bfc2fa507e1721c82651b5c1ba978a3122a",
1360
- "sha256": "9e1c2de08296f90b4f61849b10330bfc2fa507e1721c82651b5c1ba978a3122a",
1361
- "source_bytes": 115839947,
1362
- "bytes": 115839947,
1363
- "replacements": {},
1364
- "rows": 585
1365
- },
1366
- {
1367
- "path": "coderank/dumps/summarize_definitions-model-requests.jsonl",
1368
- "source_sha256": "f3d9b86b2cd59f2a93eee8f9aecf4c18472ec7d571f278da2d4f521964de0e0b",
1369
- "sha256": "f3d9b86b2cd59f2a93eee8f9aecf4c18472ec7d571f278da2d4f521964de0e0b",
1370
- "source_bytes": 21874065,
1371
- "bytes": 21874065,
1372
- "replacements": {},
1373
- "rows": 392
1374
- },
1375
- {
1376
- "path": "coderank/dumps/summarize_definitions-model-ripgrep.jsonl",
1377
- "source_sha256": "39fef2476bc1d40a4a220ebafe383d5f332d4ea1399abf3eb9d405d3f5a46f28",
1378
- "sha256": "39fef2476bc1d40a4a220ebafe383d5f332d4ea1399abf3eb9d405d3f5a46f28",
1379
- "source_bytes": 49565566,
1380
- "bytes": 49565566,
1381
- "replacements": {},
1382
- "rows": 737
1383
- },
1384
- {
1385
- "path": "coderank/dumps/summarize_definitions-model-sinatra.jsonl",
1386
- "source_sha256": "6d81a9d707e642eb1baacf99795c39278f8fb214b740a444a8f1fd7315d2880f",
1387
- "sha256": "6d81a9d707e642eb1baacf99795c39278f8fb214b740a444a8f1fd7315d2880f",
1388
- "source_bytes": 27992413,
1389
- "bytes": 27992413,
1390
- "replacements": {},
1391
- "rows": 527
1392
- },
1393
- {
1394
- "path": "coderank/dumps/summarize_definitions-model-slim.jsonl",
1395
- "source_sha256": "a40ee18ef6b1765728a7931150df5d58727204643518131f39868cb8f33dced6",
1396
- "sha256": "a40ee18ef6b1765728a7931150df5d58727204643518131f39868cb8f33dced6",
1397
- "source_bytes": 18738296,
1398
- "bytes": 18738296,
1399
- "replacements": {},
1400
- "rows": 389
1401
- },
1402
- {
1403
- "path": "coderank/dumps/summarize_definitions-model-thor.jsonl",
1404
- "source_sha256": "d53bbde3d99140eca323870fee03a9c8e23f3fe1c63c83fc3953fef283faeb66",
1405
- "sha256": "d53bbde3d99140eca323870fee03a9c8e23f3fe1c63c83fc3953fef283faeb66",
1406
- "source_bytes": 30150426,
1407
- "bytes": 30150426,
1408
- "replacements": {},
1409
- "rows": 505
1410
- },
1411
- {
1412
- "path": "coderank/dumps/summarize_definitions-model-zod.jsonl",
1413
- "source_sha256": "c77c2de2e51eaf60cccbdba6c80e6336d45e92bf28f916a39c1793089d092704",
1414
- "sha256": "c77c2de2e51eaf60cccbdba6c80e6336d45e92bf28f916a39c1793089d092704",
1415
- "source_bytes": 312916454,
1416
- "bytes": 312916454,
1417
- "replacements": {},
1418
- "rows": 926
1419
- },
1420
- {
1421
- "path": "coderank/dumps/summarize_structure-docs-axios.jsonl",
1422
- "source_sha256": "982a662c6c228be3d6411f9935015fbd2f209cb07b67a59a2af1e59c3180f0d8",
1423
- "sha256": "982a662c6c228be3d6411f9935015fbd2f209cb07b67a59a2af1e59c3180f0d8",
1424
- "source_bytes": 82343,
1425
- "bytes": 82343,
1426
- "replacements": {},
1427
- "rows": 9
1428
- },
1429
- {
1430
- "path": "coderank/dumps/summarize_structure-docs-axum.jsonl",
1431
- "source_sha256": "e463f81df1240c9541e9058e52e2d790b56745f4f2def695e4d4ebb9b847f8c8",
1432
- "sha256": "e463f81df1240c9541e9058e52e2d790b56745f4f2def695e4d4ebb9b847f8c8",
1433
- "source_bytes": 318577,
1434
- "bytes": 318577,
1435
- "replacements": {},
1436
- "rows": 53
1437
- },
1438
- {
1439
- "path": "coderank/dumps/summarize_structure-docs-cobra.jsonl",
1440
- "source_sha256": "05dc674b80797d26206444eb61e0e89837037ad66b7a44543537781b03c2e481",
1441
- "sha256": "05dc674b80797d26206444eb61e0e89837037ad66b7a44543537781b03c2e481",
1442
- "source_bytes": 516625,
1443
- "bytes": 516625,
1444
- "replacements": {},
1445
- "rows": 33
1446
- },
1447
- {
1448
- "path": "coderank/dumps/summarize_structure-docs-fd.jsonl",
1449
- "source_sha256": "0907291d079193c4bdcf1a15e189c287deeaa2cc5710eaabdbe1525e0d301eef",
1450
- "sha256": "0907291d079193c4bdcf1a15e189c287deeaa2cc5710eaabdbe1525e0d301eef",
1451
- "source_bytes": 110639,
1452
- "bytes": 110639,
1453
- "replacements": {},
1454
- "rows": 15
1455
- },
1456
- {
1457
- "path": "coderank/dumps/summarize_structure-docs-gin.jsonl",
1458
- "source_sha256": "bf7d2e245155657138e1e7d1bcfed2efe861d9c57a114d8ad47702d69595a7a0",
1459
- "sha256": "bf7d2e245155657138e1e7d1bcfed2efe861d9c57a114d8ad47702d69595a7a0",
1460
- "source_bytes": 1780676,
1461
- "bytes": 1780676,
1462
- "replacements": {},
1463
- "rows": 72
1464
- },
1465
- {
1466
- "path": "coderank/dumps/summarize_structure-docs-gson.jsonl",
1467
- "source_sha256": "01036e8afcfa61bb268a532ccf6d66030076411a8f12eda15d429fcb608e899b",
1468
- "sha256": "01036e8afcfa61bb268a532ccf6d66030076411a8f12eda15d429fcb608e899b",
1469
- "source_bytes": 1830460,
1470
- "bytes": 1830460,
1471
- "replacements": {},
1472
- "rows": 73
1473
- },
1474
- {
1475
- "path": "coderank/dumps/summarize_structure-docs-httpx.jsonl",
1476
- "source_sha256": "c4f2023db3d380c01f145fcde2e25a5f9b50371ab7178924360abc15cada2c45",
1477
- "sha256": "c4f2023db3d380c01f145fcde2e25a5f9b50371ab7178924360abc15cada2c45",
1478
- "source_bytes": 9149,
1479
- "bytes": 9149,
1480
- "replacements": {},
1481
- "rows": 1
1482
- },
1483
- {
1484
- "path": "coderank/dumps/summarize_structure-docs-okhttp.jsonl",
1485
- "source_sha256": "ef4a91f7ce9d3c5b77fd843e11e85b79e6ae64e2728a0f5270b00e17f1ae2b05",
1486
- "sha256": "ef4a91f7ce9d3c5b77fd843e11e85b79e6ae64e2728a0f5270b00e17f1ae2b05",
1487
- "source_bytes": 3561853,
1488
- "bytes": 3561853,
1489
- "replacements": {},
1490
- "rows": 184
1491
- },
1492
- {
1493
- "path": "coderank/dumps/summarize_structure-docs-ripgrep.jsonl",
1494
- "source_sha256": "569e705c1f9b9a97ece5bdd80b7a97625a22200eca76bef3ba2d0e891960f453",
1495
- "sha256": "569e705c1f9b9a97ece5bdd80b7a97625a22200eca76bef3ba2d0e891960f453",
1496
- "source_bytes": 1033696,
1497
- "bytes": 1033696,
1498
- "replacements": {},
1499
- "rows": 243
1500
- },
1501
- {
1502
- "path": "coderank/dumps/summarize_structure-docs-sinatra.jsonl",
1503
- "source_sha256": "43da310090a2d9991ceb05f9d478b714e2d0c64203f33a935d56d70eb3b3e6bc",
1504
- "sha256": "43da310090a2d9991ceb05f9d478b714e2d0c64203f33a935d56d70eb3b3e6bc",
1505
- "source_bytes": 284407,
1506
- "bytes": 284407,
1507
- "replacements": {},
1508
- "rows": 29
1509
- },
1510
- {
1511
- "path": "coderank/dumps/summarize_structure-docs-slim.jsonl",
1512
- "source_sha256": "3d45525214e66953314f5afc8450c6fdd50b391cca4816582344bcc155633e94",
1513
- "sha256": "3d45525214e66953314f5afc8450c6fdd50b391cca4816582344bcc155633e94",
1514
- "source_bytes": 101402,
1515
- "bytes": 101402,
1516
- "replacements": {},
1517
- "rows": 10
1518
- },
1519
- {
1520
- "path": "coderank/dumps/summarize_structure-docs-thor.jsonl",
1521
- "source_sha256": "e02745dc7af454543d83c8b1955077d57e3ebb0247e52fee70aacdab2c4b0281",
1522
- "sha256": "e02745dc7af454543d83c8b1955077d57e3ebb0247e52fee70aacdab2c4b0281",
1523
- "source_bytes": 224738,
1524
- "bytes": 224738,
1525
- "replacements": {},
1526
- "rows": 24
1527
- },
1528
- {
1529
- "path": "coderank/dumps/summarize_structure-docs-zod.jsonl",
1530
- "source_sha256": "9607afc3ee04a1fdd2600fea082730ee990c7e3ebc24e8f8d7aab13c7ae0841e",
1531
- "sha256": "9607afc3ee04a1fdd2600fea082730ee990c7e3ebc24e8f8d7aab13c7ae0841e",
1532
- "source_bytes": 5166911,
1533
- "bytes": 5166911,
1534
- "replacements": {},
1535
- "rows": 187
1536
- },
1537
- {
1538
- "path": "coderank/dumps/summarize_structure-model-axios.jsonl",
1539
- "source_sha256": "29d39e366c4cdf9e0b328eef1af4de50c55571471fbf0eeb3d72f895013f1466",
1540
- "sha256": "29d39e366c4cdf9e0b328eef1af4de50c55571471fbf0eeb3d72f895013f1466",
1541
- "source_bytes": 5335485,
1542
- "bytes": 5335485,
1543
- "replacements": {},
1544
- "rows": 583
1545
- },
1546
- {
1547
- "path": "coderank/dumps/summarize_structure-model-axum.jsonl",
1548
- "source_sha256": "022a604b69e6e8c63cf9d1ab5d55f5073b5c8c6ed6d05ef9a2411f303c5d2794",
1549
- "sha256": "022a604b69e6e8c63cf9d1ab5d55f5073b5c8c6ed6d05ef9a2411f303c5d2794",
1550
- "source_bytes": 4413531,
1551
- "bytes": 4413531,
1552
- "replacements": {},
1553
- "rows": 674
1554
- },
1555
- {
1556
- "path": "coderank/dumps/summarize_structure-model-cobra.jsonl",
1557
- "source_sha256": "f8d572042f14a9a9946e2933f2a6beab51c993c8f4b26179fa83f75e63e53190",
1558
- "sha256": "f8d572042f14a9a9946e2933f2a6beab51c993c8f4b26179fa83f75e63e53190",
1559
- "source_bytes": 4196514,
1560
- "bytes": 4196514,
1561
- "replacements": {},
1562
- "rows": 287
1563
- },
1564
- {
1565
- "path": "coderank/dumps/summarize_structure-model-express.jsonl",
1566
- "source_sha256": "b534e0995e7f0781665787b2086fd50c2a4a45873cf63d7c297f1e828c357328",
1567
- "sha256": "b534e0995e7f0781665787b2086fd50c2a4a45873cf63d7c297f1e828c357328",
1568
- "source_bytes": 785302,
1569
- "bytes": 785302,
1570
- "replacements": {},
1571
- "rows": 98
1572
- },
1573
- {
1574
- "path": "coderank/dumps/summarize_structure-model-fd.jsonl",
1575
- "source_sha256": "60d21005597530635b01c020dd308acf6c5464e74affc67e33f69654d0af6770",
1576
- "sha256": "60d21005597530635b01c020dd308acf6c5464e74affc67e33f69654d0af6770",
1577
- "source_bytes": 1215476,
1578
- "bytes": 1215476,
1579
- "replacements": {},
1580
- "rows": 214
1581
- },
1582
- {
1583
- "path": "coderank/dumps/summarize_structure-model-flask.jsonl",
1584
- "source_sha256": "cf121de2c1493634573e67e86668e0541448a3366da6b1afc15689d80c7d7034",
1585
- "sha256": "cf121de2c1493634573e67e86668e0541448a3366da6b1afc15689d80c7d7034",
1586
- "source_bytes": 8532997,
1587
- "bytes": 8532997,
1588
- "replacements": {},
1589
- "rows": 779
1590
- },
1591
- {
1592
- "path": "coderank/dumps/summarize_structure-model-gin.jsonl",
1593
- "source_sha256": "d3ebc36a3be4ce4685245ba8f11c5e46401a6c2ae9cf905117218bcc412d3546",
1594
- "sha256": "d3ebc36a3be4ce4685245ba8f11c5e46401a6c2ae9cf905117218bcc412d3546",
1595
- "source_bytes": 16633104,
1596
- "bytes": 16633104,
1597
- "replacements": {},
1598
- "rows": 702
1599
- },
1600
- {
1601
- "path": "coderank/dumps/summarize_structure-model-gson.jsonl",
1602
- "source_sha256": "d3795241fa1c3445466775cfcf6ac86b559825454cc2e64ac04da9e9efec82ca",
1603
- "sha256": "d3795241fa1c3445466775cfcf6ac86b559825454cc2e64ac04da9e9efec82ca",
1604
- "source_bytes": 13646774,
1605
- "bytes": 13646774,
1606
- "replacements": {},
1607
- "rows": 584
1608
- },
1609
- {
1610
- "path": "coderank/dumps/summarize_structure-model-httpx.jsonl",
1611
- "source_sha256": "f824225456e48888fb39aa238108e6a06cf1786a5cb405ad7c5108fe84cd41fd",
1612
- "sha256": "f824225456e48888fb39aa238108e6a06cf1786a5cb405ad7c5108fe84cd41fd",
1613
- "source_bytes": 4391684,
1614
- "bytes": 4391684,
1615
- "replacements": {},
1616
- "rows": 547
1617
- },
1618
- {
1619
- "path": "coderank/dumps/summarize_structure-model-okhttp.jsonl",
1620
- "source_sha256": "df04fb09c053ee6b20501dacfe2829a86f0eb3fae9b58b6efd3939844db2e382",
1621
- "sha256": "df04fb09c053ee6b20501dacfe2829a86f0eb3fae9b58b6efd3939844db2e382",
1622
- "source_bytes": 11859794,
1623
- "bytes": 11859794,
1624
- "replacements": {},
1625
- "rows": 585
1626
- },
1627
- {
1628
- "path": "coderank/dumps/summarize_structure-model-requests.jsonl",
1629
- "source_sha256": "f2aca8248d0dfa6b2951e3d6027043898e0262a1a5f79221b7d48c7aaca17222",
1630
- "sha256": "f2aca8248d0dfa6b2951e3d6027043898e0262a1a5f79221b7d48c7aaca17222",
1631
- "source_bytes": 3019106,
1632
- "bytes": 3019106,
1633
- "replacements": {},
1634
- "rows": 392
1635
- },
1636
- {
1637
- "path": "coderank/dumps/summarize_structure-model-ripgrep.jsonl",
1638
- "source_sha256": "d2e9cedc4a2ee47f78bdde1d61bc30faa1766adc8d51a9dd7d9b83dcb2725830",
1639
- "sha256": "d2e9cedc4a2ee47f78bdde1d61bc30faa1766adc8d51a9dd7d9b83dcb2725830",
1640
- "source_bytes": 2858716,
1641
- "bytes": 2858716,
1642
- "replacements": {},
1643
- "rows": 737
1644
- },
1645
- {
1646
- "path": "coderank/dumps/summarize_structure-model-sinatra.jsonl",
1647
- "source_sha256": "2733680220b158e7500c411cebdeeb03e38c98747528b22079e629dc6d2cd9e4",
1648
- "sha256": "2733680220b158e7500c411cebdeeb03e38c98747528b22079e629dc6d2cd9e4",
1649
- "source_bytes": 6181624,
1650
- "bytes": 6181624,
1651
- "replacements": {},
1652
- "rows": 527
1653
- },
1654
- {
1655
- "path": "coderank/dumps/summarize_structure-model-slim.jsonl",
1656
- "source_sha256": "ce1fe65ce345e6dee0de309b140c8dcd9aa857ba028e307864a21f0762375398",
1657
- "sha256": "ce1fe65ce345e6dee0de309b140c8dcd9aa857ba028e307864a21f0762375398",
1658
- "source_bytes": 4184002,
1659
- "bytes": 4184002,
1660
- "replacements": {},
1661
- "rows": 389
1662
- },
1663
- {
1664
- "path": "coderank/dumps/summarize_structure-model-thor.jsonl",
1665
- "source_sha256": "b272291b1f0e4b0aa4fca80d5675918dea18814a33d3568de9e83376c5e701df",
1666
- "sha256": "b272291b1f0e4b0aa4fca80d5675918dea18814a33d3568de9e83376c5e701df",
1667
- "source_bytes": 4603762,
1668
- "bytes": 4603762,
1669
- "replacements": {},
1670
- "rows": 505
1671
- },
1672
- {
1673
- "path": "coderank/dumps/summarize_structure-model-zod.jsonl",
1674
- "source_sha256": "6b5e7678c24d7fd7175802a5608a468916f17e02773a33a02660c50ee58867b0",
1675
- "sha256": "6b5e7678c24d7fd7175802a5608a468916f17e02773a33a02660c50ee58867b0",
1676
- "source_bytes": 25383996,
1677
- "bytes": 25383996,
1678
- "replacements": {},
1679
- "rows": 926
1680
- },
1681
- {
1682
- "path": "coderank/sites/rows-sites-axios.jsonl",
1683
- "source_sha256": "b33046f69b6d1951f5389c0720f5b0ce8687228515dbc688a4b8f6b7e7bdd7a0",
1684
- "sha256": "b33046f69b6d1951f5389c0720f5b0ce8687228515dbc688a4b8f6b7e7bdd7a0",
1685
- "source_bytes": 12478914,
1686
- "bytes": 12478914,
1687
- "replacements": {},
1688
- "rows": 2238
1689
- },
1690
- {
1691
- "path": "coderank/sites/rows-sites-axum.jsonl",
1692
- "source_sha256": "540be4c128810f3a0e85cf0b1b71b05a55a32715692715d83a54892ac20c6c4d",
1693
- "sha256": "540be4c128810f3a0e85cf0b1b71b05a55a32715692715d83a54892ac20c6c4d",
1694
- "source_bytes": 18768534,
1695
- "bytes": 18768534,
1696
- "replacements": {},
1697
- "rows": 3147
1698
- },
1699
- {
1700
- "path": "coderank/sites/rows-sites-cobra.jsonl",
1701
- "source_sha256": "9b0da6f328d72497abbed61161de1a429664579c388f31c8e6ff91ed939f06ec",
1702
- "sha256": "9b0da6f328d72497abbed61161de1a429664579c388f31c8e6ff91ed939f06ec",
1703
- "source_bytes": 5317822,
1704
- "bytes": 5317822,
1705
- "replacements": {},
1706
- "rows": 1138
1707
- },
1708
- {
1709
- "path": "coderank/sites/rows-sites-express.jsonl",
1710
- "source_sha256": "cf6563e0af829f92bb106f1a5fd196a97711902d90b6570e530c55c9b9f8b889",
1711
- "sha256": "cf6563e0af829f92bb106f1a5fd196a97711902d90b6570e530c55c9b9f8b889",
1712
- "source_bytes": 1481708,
1713
- "bytes": 1481708,
1714
- "replacements": {},
1715
- "rows": 315
1716
- },
1717
- {
1718
- "path": "coderank/sites/rows-sites-fd.jsonl",
1719
- "source_sha256": "609d47a13956a1c928fb3185190146ab3ea0a89afb5aba26c20044e7d77775d8",
1720
- "sha256": "609d47a13956a1c928fb3185190146ab3ea0a89afb5aba26c20044e7d77775d8",
1721
- "source_bytes": 4468124,
1722
- "bytes": 4468124,
1723
- "replacements": {},
1724
- "rows": 933
1725
- },
1726
- {
1727
- "path": "coderank/sites/rows-sites-flask.jsonl",
1728
- "source_sha256": "cf4de7b99baa7dcbacf26e337d583cba29433678133d2f6540e85707e788f22e",
1729
- "sha256": "cf4de7b99baa7dcbacf26e337d583cba29433678133d2f6540e85707e788f22e",
1730
- "source_bytes": 15285593,
1731
- "bytes": 15285593,
1732
- "replacements": {},
1733
- "rows": 3077
1734
- },
1735
- {
1736
- "path": "coderank/sites/rows-sites-gin.jsonl",
1737
- "source_sha256": "12d4d08025732b9ee5e0a2bdd7d0ce906c98d0732e03ff3f1cbee0168a939224",
1738
- "sha256": "12d4d08025732b9ee5e0a2bdd7d0ce906c98d0732e03ff3f1cbee0168a939224",
1739
- "source_bytes": 13328763,
1740
- "bytes": 13328763,
1741
- "replacements": {},
1742
- "rows": 2784
1743
- },
1744
- {
1745
- "path": "coderank/sites/rows-sites-gson.jsonl",
1746
- "source_sha256": "9f45410c3e68e187d636eabd2aa195286b5bf01e33b01f63fbacd683c8c92389",
1747
- "sha256": "9f45410c3e68e187d636eabd2aa195286b5bf01e33b01f63fbacd683c8c92389",
1748
- "source_bytes": 19109661,
1749
- "bytes": 19109661,
1750
- "replacements": {},
1751
- "rows": 2466
1752
- },
1753
- {
1754
- "path": "coderank/sites/rows-sites-httpx.jsonl",
1755
- "source_sha256": "45c13cc83c54419f1e57bbedac4377dddb84b1c779e1c838634bc01787365be2",
1756
- "sha256": "45c13cc83c54419f1e57bbedac4377dddb84b1c779e1c838634bc01787365be2",
1757
- "source_bytes": 10667622,
1758
- "bytes": 10667622,
1759
- "replacements": {},
1760
- "rows": 2182
1761
- },
1762
- {
1763
- "path": "coderank/sites/rows-sites-okhttp.jsonl",
1764
- "source_sha256": "37ad7a288e1d0e8fe92dfba0ed0181e7b0afd56b5104aebd208d99d127028e6b",
1765
- "sha256": "37ad7a288e1d0e8fe92dfba0ed0181e7b0afd56b5104aebd208d99d127028e6b",
1766
- "source_bytes": 18278864,
1767
- "bytes": 18278864,
1768
- "replacements": {},
1769
- "rows": 2432
1770
- },
1771
- {
1772
- "path": "coderank/sites/rows-sites-requests.jsonl",
1773
- "source_sha256": "1c2964d79f32b8494bcb76d4f07a93dbb130dcfe2278440d536c9b78f7c599ab",
1774
- "sha256": "1c2964d79f32b8494bcb76d4f07a93dbb130dcfe2278440d536c9b78f7c599ab",
1775
- "source_bytes": 7502945,
1776
- "bytes": 7502945,
1777
- "replacements": {},
1778
- "rows": 1521
1779
- },
1780
- {
1781
- "path": "coderank/sites/rows-sites-slim.jsonl",
1782
- "source_sha256": "7d5fa209fcb688d9b05b02aa13503869d5f63a3517fbe463f21f88ae05c85cd8",
1783
- "sha256": "7d5fa209fcb688d9b05b02aa13503869d5f63a3517fbe463f21f88ae05c85cd8",
1784
- "source_bytes": 9096327,
1785
- "bytes": 9096327,
1786
- "replacements": {},
1787
- "rows": 1568
1788
- },
1789
- {
1790
- "path": "coderank/sites/rows-sites-thor.jsonl",
1791
- "source_sha256": "aae839ef2eeb2188c85015f338cc0ed74798dc925b5118c54c1fd45b3e843fce",
1792
- "sha256": "aae839ef2eeb2188c85015f338cc0ed74798dc925b5118c54c1fd45b3e843fce",
1793
- "source_bytes": 9671452,
1794
- "bytes": 9671452,
1795
- "replacements": {},
1796
- "rows": 1876
1797
- },
1798
- {
1799
- "path": "licenses/axios/LICENSE",
1800
- "source_sha256": "82761059eaedacb3356803aea8a170d8298609f91b14fc32ee1bfb40d690183c",
1801
- "sha256": "82761059eaedacb3356803aea8a170d8298609f91b14fc32ee1bfb40d690183c",
1802
- "source_bytes": 1084,
1803
- "bytes": 1084,
1804
- "replacements": {}
1805
- },
1806
- {
1807
- "path": "licenses/axum/LICENSE",
1808
- "source_sha256": "008c87afcd2e626eaf564093250bed06dd7efb5732113264bba3dda8f1c556a1",
1809
- "sha256": "008c87afcd2e626eaf564093250bed06dd7efb5732113264bba3dda8f1c556a1",
1810
- "source_bytes": 1080,
1811
- "bytes": 1080,
1812
- "replacements": {}
1813
- },
1814
- {
1815
- "path": "licenses/cobra/LICENSE.txt",
1816
- "source_sha256": "5e3400b93bbb099e83e52bab885e7441750673c21f97988ca3f1240639b63283",
1817
- "sha256": "5e3400b93bbb099e83e52bab885e7441750673c21f97988ca3f1240639b63283",
1818
- "source_bytes": 10140,
1819
- "bytes": 10140,
1820
- "replacements": {}
1821
- },
1822
- {
1823
- "path": "licenses/express/LICENSE",
1824
- "source_sha256": "95a5762890e5c1c9808921cef095661fc482c5e1f0bba31446ac85595df6237c",
1825
- "sha256": "95a5762890e5c1c9808921cef095661fc482c5e1f0bba31446ac85595df6237c",
1826
- "source_bytes": 1249,
1827
- "bytes": 1249,
1828
- "replacements": {}
1829
- },
1830
- {
1831
- "path": "licenses/fd/LICENSE-APACHE",
1832
- "source_sha256": "73c83c60d817e7df1943cb3f0af81e4939a8352c9a96c2fd00451b1116fa635c",
1833
- "sha256": "73c83c60d817e7df1943cb3f0af81e4939a8352c9a96c2fd00451b1116fa635c",
1834
- "source_bytes": 10838,
1835
- "bytes": 10838,
1836
- "replacements": {}
1837
- },
1838
- {
1839
- "path": "licenses/fd/LICENSE-MIT",
1840
- "source_sha256": "322cfc7aa0c774d0eca3b2610f1d414de3ddbd7d8dd4b9dea941a13a6eb07455",
1841
- "sha256": "322cfc7aa0c774d0eca3b2610f1d414de3ddbd7d8dd4b9dea941a13a6eb07455",
1842
- "source_bytes": 1082,
1843
- "bytes": 1082,
1844
- "replacements": {}
1845
- },
1846
- {
1847
- "path": "licenses/flask/LICENSE.txt",
1848
- "source_sha256": "489a8e1108509ed98a37bb983e11e0f7e1d31f0bd8f99a79c8448e7ff37d07ea",
1849
- "sha256": "489a8e1108509ed98a37bb983e11e0f7e1d31f0bd8f99a79c8448e7ff37d07ea",
1850
- "source_bytes": 1475,
1851
- "bytes": 1475,
1852
- "replacements": {}
1853
- },
1854
- {
1855
- "path": "licenses/gin/LICENSE",
1856
- "source_sha256": "b104efb2c7700691650f27034e8541c5ae0ed9af54d884287f19a7467ca2fe7f",
1857
- "sha256": "b104efb2c7700691650f27034e8541c5ae0ed9af54d884287f19a7467ca2fe7f",
1858
- "source_bytes": 1099,
1859
- "bytes": 1099,
1860
- "replacements": {}
1861
- },
1862
- {
1863
- "path": "licenses/gson/LICENSE",
1864
- "source_sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
1865
- "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
1866
- "source_bytes": 11358,
1867
- "bytes": 11358,
1868
- "replacements": {}
1869
- },
1870
- {
1871
- "path": "licenses/httpx/LICENSE.md",
1872
- "source_sha256": "4ec59d544f12b5f539a3a716fd321ac58ccd8030b465221f2c880200cdf28d8d",
1873
- "sha256": "4ec59d544f12b5f539a3a716fd321ac58ccd8030b465221f2c880200cdf28d8d",
1874
- "source_bytes": 1508,
1875
- "bytes": 1508,
1876
- "replacements": {}
1877
- },
1878
- {
1879
- "path": "licenses/okhttp/LICENSE.txt",
1880
- "source_sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
1881
- "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
1882
- "source_bytes": 11358,
1883
- "bytes": 11358,
1884
- "replacements": {}
1885
- },
1886
- {
1887
- "path": "licenses/requests/LICENSE",
1888
- "source_sha256": "09e8a9bcec8067104652c168685ab0931e7868f9c8284b66f5ae6edae5f1130b",
1889
- "sha256": "09e8a9bcec8067104652c168685ab0931e7868f9c8284b66f5ae6edae5f1130b",
1890
- "source_bytes": 10142,
1891
- "bytes": 10142,
1892
- "replacements": {}
1893
- },
1894
- {
1895
- "path": "licenses/requests/NOTICE",
1896
- "source_sha256": "f5110972dedad2b4e9d314518daf3b7d72d6e02e499acd802181de6f74571dcc",
1897
- "sha256": "f5110972dedad2b4e9d314518daf3b7d72d6e02e499acd802181de6f74571dcc",
1898
- "source_bytes": 38,
1899
- "bytes": 38,
1900
- "replacements": {}
1901
- },
1902
- {
1903
- "path": "licenses/ripgrep/COPYING",
1904
- "source_sha256": "01c266bced4a434da0051174d6bee16a4c82cf634e2679b6155d40d75012390f",
1905
- "sha256": "01c266bced4a434da0051174d6bee16a4c82cf634e2679b6155d40d75012390f",
1906
- "source_bytes": 126,
1907
- "bytes": 126,
1908
- "replacements": {}
1909
- },
1910
- {
1911
- "path": "licenses/ripgrep/LICENSE-MIT",
1912
- "source_sha256": "0f96a83840e146e43c0ec96a22ec1f392e0680e6c1226e6f3ba87e0740af850f",
1913
- "sha256": "0f96a83840e146e43c0ec96a22ec1f392e0680e6c1226e6f3ba87e0740af850f",
1914
- "source_bytes": 1081,
1915
- "bytes": 1081,
1916
- "replacements": {}
1917
- },
1918
- {
1919
- "path": "licenses/ripgrep/UNLICENSE",
1920
- "source_sha256": "7e12e5df4bae12cb21581ba157ced20e1986a0508dd10d0e8a4ab9a4cf94e85c",
1921
- "sha256": "7e12e5df4bae12cb21581ba157ced20e1986a0508dd10d0e8a4ab9a4cf94e85c",
1922
- "source_bytes": 1211,
1923
- "bytes": 1211,
1924
- "replacements": {}
1925
- },
1926
- {
1927
- "path": "licenses/sinatra/LICENSE",
1928
- "source_sha256": "f2f60556a2d5c8d7ef9d54ec1c62152ff40d626b807aba5bdc5126f669ed500f",
1929
- "sha256": "f2f60556a2d5c8d7ef9d54ec1c62152ff40d626b807aba5bdc5126f669ed500f",
1930
- "source_bytes": 1172,
1931
- "bytes": 1172,
1932
- "replacements": {}
1933
- },
1934
- {
1935
- "path": "licenses/slim/LICENSE.md",
1936
- "source_sha256": "24d84675cc9fffd507101fd43fecb5dd7f87679f065d90b2ef597d0a8b7e5994",
1937
- "sha256": "24d84675cc9fffd507101fd43fecb5dd7f87679f065d90b2ef597d0a8b7e5994",
1938
- "source_bytes": 1062,
1939
- "bytes": 1062,
1940
- "replacements": {}
1941
- },
1942
- {
1943
- "path": "licenses/thor/LICENSE.md",
1944
- "source_sha256": "4df3e92a124b84d6767b1a0f1302600ea41d7616cda0b0fc40c1eefe4c9e414b",
1945
- "sha256": "4df3e92a124b84d6767b1a0f1302600ea41d7616cda0b0fc40c1eefe4c9e414b",
1946
- "source_bytes": 1075,
1947
- "bytes": 1075,
1948
- "replacements": {}
1949
- },
1950
- {
1951
- "path": "licenses/zod/LICENSE",
1952
- "source_sha256": "3f1189b28e3866e0d979968d466b78f813f76827cfdca1fbb124cc0a5c8841f8",
1953
- "sha256": "3f1189b28e3866e0d979968d466b78f813f76827cfdca1fbb124cc0a5c8841f8",
1954
- "source_bytes": 1072,
1955
- "bytes": 1072,
1956
- "replacements": {}
1957
- },
1958
- {
1959
- "path": "LICENSE",
1960
- "source_sha256": "3c2307bbbbb88f79560a165b098c600da8f20fb1ec024586d4be192ebc9c6f8e",
1961
- "sha256": "3c2307bbbbb88f79560a165b098c600da8f20fb1ec024586d4be192ebc9c6f8e",
1962
- "source_bytes": 12099,
1963
- "bytes": 12099,
1964
- "replacements": {}
1965
- },
1966
- {
1967
- "path": "training/code-rank/scripts/decide/bench_latency.py",
1968
- "source_sha256": "4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f",
1969
- "sha256": "4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f",
1970
- "source_bytes": 3094,
1971
- "bytes": 3094,
1972
- "replacements": {}
1973
- },
1974
- {
1975
- "path": "training/code-rank/scripts/decide/calibrate.py",
1976
- "source_sha256": "c4594e1514fffb0b64b931476c37890ce872259babb28aeaadc934e737a8c607",
1977
- "sha256": "c4594e1514fffb0b64b931476c37890ce872259babb28aeaadc934e737a8c607",
1978
- "source_bytes": 9284,
1979
- "bytes": 9284,
1980
- "replacements": {}
1981
- },
1982
- {
1983
- "path": "training/code-rank/scripts/decide/corpus.py",
1984
- "source_sha256": "d0c7dda2d181bb81646e61090a971ebc1c5922703f00170572ed2ce40ce8652b",
1985
- "sha256": "d0c7dda2d181bb81646e61090a971ebc1c5922703f00170572ed2ce40ce8652b",
1986
- "source_bytes": 5921,
1987
- "bytes": 5921,
1988
- "replacements": {}
1989
- },
1990
- {
1991
- "path": "training/code-rank/scripts/decide/headfile.py",
1992
- "source_sha256": "f4ca003205839d391c10d26417af8239b86b8af8471f30969441346edcf770b5",
1993
- "sha256": "f4ca003205839d391c10d26417af8239b86b8af8471f30969441346edcf770b5",
1994
- "source_bytes": 2504,
1995
- "bytes": 2504,
1996
- "replacements": {}
1997
- },
1998
- {
1999
- "path": "training/code-rank/scripts/decide/parity_probe.py",
2000
- "source_sha256": "b574de47f0deed015f59d71cbfd50130eaf6a40c9a54c3de261a66d6b0f4df70",
2001
- "sha256": "b574de47f0deed015f59d71cbfd50130eaf6a40c9a54c3de261a66d6b0f4df70",
2002
- "source_bytes": 4302,
2003
- "bytes": 4302,
2004
- "replacements": {}
2005
- },
2006
- {
2007
- "path": "training/code-rank/scripts/decide/rank-rubric.json",
2008
- "source_sha256": "f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98",
2009
- "sha256": "f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98",
2010
- "source_bytes": 139,
2011
- "bytes": 139,
2012
- "replacements": {}
2013
- },
2014
- {
2015
- "path": "training/code-rank/scripts/decide/replay_eval.py",
2016
- "source_sha256": "4a2dc07a0984294030c7691c9727ce98392e8ef578fbfa491ae7945635577172",
2017
- "sha256": "4a2dc07a0984294030c7691c9727ce98392e8ef578fbfa491ae7945635577172",
2018
- "source_bytes": 18668,
2019
- "bytes": 18668,
2020
- "replacements": {}
2021
- },
2022
- {
2023
- "path": "training/code-rank/scripts/decide/rerank/synthesize_sites.py",
2024
- "source_sha256": "60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d",
2025
- "sha256": "60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d",
2026
- "source_bytes": 15544,
2027
- "bytes": 15544,
2028
- "replacements": {}
2029
- },
2030
- {
2031
- "path": "training/code-rank/scripts/decide/rerank/train_rerank.py",
2032
- "source_sha256": "b071fa53ed417afc7c8a1779460d60950d30f8472daf4ef4eb20a976815e6f9a",
2033
- "sha256": "b071fa53ed417afc7c8a1779460d60950d30f8472daf4ef4eb20a976815e6f9a",
2034
- "source_bytes": 9484,
2035
- "bytes": 9484,
2036
- "replacements": {}
2037
- },
2038
- {
2039
- "path": "training/code-rank/scripts/decide/row.schema.json",
2040
- "source_sha256": "260f26ccdd53437f1d01b269d9f510dde1b497b9b64a45a7807d83231a2a7bb5",
2041
- "sha256": "260f26ccdd53437f1d01b269d9f510dde1b497b9b64a45a7807d83231a2a7bb5",
2042
- "source_bytes": 6241,
2043
- "bytes": 6241,
2044
- "replacements": {}
2045
- },
2046
- {
2047
- "path": "training/code-rank/scripts/decide/rows.py",
2048
- "source_sha256": "9d9ec7c18661ffa53820f272b2324fec4a1d72ee578ca5f0df22ae5bda364b17",
2049
- "sha256": "9d9ec7c18661ffa53820f272b2324fec4a1d72ee578ca5f0df22ae5bda364b17",
2050
- "source_bytes": 1359,
2051
- "bytes": 1359,
2052
- "replacements": {}
2053
- },
2054
- {
2055
- "path": "training/code-rank/scripts/decide/train.py",
2056
- "source_sha256": "e9343227428319b8c3e75bf8de2976f78fefd9f3720ce3cfdd4a3c840c456772",
2057
- "sha256": "e9343227428319b8c3e75bf8de2976f78fefd9f3720ce3cfdd4a3c840c456772",
2058
- "source_bytes": 25220,
2059
- "bytes": 25220,
2060
- "replacements": {}
2061
- },
2062
- {
2063
- "path": "training/unit-rank/scripts/decide/bench_latency.py",
2064
- "source_sha256": "4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f",
2065
- "sha256": "4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f",
2066
- "source_bytes": 3094,
2067
- "bytes": 3094,
2068
- "replacements": {}
2069
- },
2070
- {
2071
- "path": "training/unit-rank/scripts/decide/calibrate.py",
2072
- "source_sha256": "6cef23a6e54fdb048e223cb336ca0d4f32168b28190208259befdb8cfcceac2b",
2073
- "sha256": "6cef23a6e54fdb048e223cb336ca0d4f32168b28190208259befdb8cfcceac2b",
2074
- "source_bytes": 10475,
2075
- "bytes": 10475,
2076
- "replacements": {}
2077
- },
2078
- {
2079
- "path": "training/unit-rank/scripts/decide/corpus.py",
2080
- "source_sha256": "d0c7dda2d181bb81646e61090a971ebc1c5922703f00170572ed2ce40ce8652b",
2081
- "sha256": "d0c7dda2d181bb81646e61090a971ebc1c5922703f00170572ed2ce40ce8652b",
2082
- "source_bytes": 5921,
2083
- "bytes": 5921,
2084
- "replacements": {}
2085
- },
2086
- {
2087
- "path": "training/unit-rank/scripts/decide/headfile.py",
2088
- "source_sha256": "e6746d919b5aa9217d1e666b97dd09c807442c31d24fb04654dc22035a4014cd",
2089
- "sha256": "e6746d919b5aa9217d1e666b97dd09c807442c31d24fb04654dc22035a4014cd",
2090
- "source_bytes": 3414,
2091
- "bytes": 3414,
2092
- "replacements": {}
2093
- },
2094
- {
2095
- "path": "training/unit-rank/scripts/decide/parity_probe.py",
2096
- "source_sha256": "b2ea43b655b6dceb637ab7fbee9ff3755c6cf34b976cfeadd6119a0734e2a9b9",
2097
- "sha256": "b2ea43b655b6dceb637ab7fbee9ff3755c6cf34b976cfeadd6119a0734e2a9b9",
2098
- "source_bytes": 5777,
2099
- "bytes": 5777,
2100
- "replacements": {}
2101
- },
2102
- {
2103
- "path": "training/unit-rank/scripts/decide/rank-rubric.json",
2104
- "source_sha256": "f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98",
2105
- "sha256": "f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98",
2106
- "source_bytes": 139,
2107
- "bytes": 139,
2108
- "replacements": {}
2109
- },
2110
- {
2111
- "path": "training/unit-rank/scripts/decide/replay_eval.py",
2112
- "source_sha256": "9fe36ccc8021cc62666630ac0f41ecf5c1d3a0b8a24ad438461f8120a9eb6228",
2113
- "sha256": "9fe36ccc8021cc62666630ac0f41ecf5c1d3a0b8a24ad438461f8120a9eb6228",
2114
- "source_bytes": 24237,
2115
- "bytes": 24237,
2116
- "replacements": {}
2117
- },
2118
- {
2119
- "path": "training/unit-rank/scripts/decide/rerank/synthesize_sites.py",
2120
- "source_sha256": "60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d",
2121
- "sha256": "60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d",
2122
- "source_bytes": 15544,
2123
- "bytes": 15544,
2124
- "replacements": {}
2125
- },
2126
- {
2127
- "path": "training/unit-rank/scripts/decide/rerank/train_rerank.py",
2128
- "source_sha256": "1fed73f53fd2e3512ec0d626edd3b41bb0969bf52ca302dd1fc4c772737f6e59",
2129
- "sha256": "1fed73f53fd2e3512ec0d626edd3b41bb0969bf52ca302dd1fc4c772737f6e59",
2130
- "source_bytes": 9686,
2131
- "bytes": 9686,
2132
- "replacements": {}
2133
- },
2134
- {
2135
- "path": "training/unit-rank/scripts/decide/row.schema.json",
2136
- "source_sha256": "70d83cc6d22bd465cad54c696403be9bbfd7152460d74ee7ed9462046f377913",
2137
- "sha256": "70d83cc6d22bd465cad54c696403be9bbfd7152460d74ee7ed9462046f377913",
2138
- "source_bytes": 7261,
2139
- "bytes": 7261,
2140
- "replacements": {}
2141
- },
2142
- {
2143
- "path": "training/unit-rank/scripts/decide/rows.py",
2144
- "source_sha256": "2554a171844b7a432920b82a0a9a7d070fac6e7777f171e9c53b8eed557a325e",
2145
- "sha256": "2554a171844b7a432920b82a0a9a7d070fac6e7777f171e9c53b8eed557a325e",
2146
- "source_bytes": 4127,
2147
- "bytes": 4127,
2148
- "replacements": {}
2149
- },
2150
- {
2151
- "path": "training/unit-rank/scripts/decide/skill_discovery.py",
2152
- "source_sha256": "cb9bd0edd5472ba114310ba9f50a3d2089b2964696250ebea1c2ca9f0f0037e7",
2153
- "sha256": "cb9bd0edd5472ba114310ba9f50a3d2089b2964696250ebea1c2ca9f0f0037e7",
2154
- "source_bytes": 5956,
2155
- "bytes": 5956,
2156
- "replacements": {}
2157
- },
2158
- {
2159
- "path": "training/unit-rank/scripts/decide/train.py",
2160
- "source_sha256": "a70bd2179da44b6d9356585f3b86674556c85aec5cfb9b87e2958af0d180fd4f",
2161
- "sha256": "a70bd2179da44b6d9356585f3b86674556c85aec5cfb9b87e2958af0d180fd4f",
2162
- "source_bytes": 28887,
2163
- "bytes": 28887,
2164
- "replacements": {}
2165
- },
2166
- {
2167
- "path": "training/turn-load/scripts/decide/bench_latency.py",
2168
- "source_sha256": "4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f",
2169
- "sha256": "4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f",
2170
- "source_bytes": 3094,
2171
- "bytes": 3094,
2172
- "replacements": {}
2173
- },
2174
- {
2175
- "path": "training/turn-load/scripts/decide/calibrate.py",
2176
- "source_sha256": "6cef23a6e54fdb048e223cb336ca0d4f32168b28190208259befdb8cfcceac2b",
2177
- "sha256": "6cef23a6e54fdb048e223cb336ca0d4f32168b28190208259befdb8cfcceac2b",
2178
- "source_bytes": 10475,
2179
- "bytes": 10475,
2180
- "replacements": {}
2181
- },
2182
- {
2183
- "path": "training/turn-load/scripts/decide/corpus.py",
2184
- "source_sha256": "a8b8295261f19dda992c81fdec8aab8fbc2c065bc28cd186f03def6c771c7e8c",
2185
- "sha256": "a8b8295261f19dda992c81fdec8aab8fbc2c065bc28cd186f03def6c771c7e8c",
2186
- "source_bytes": 6028,
2187
- "bytes": 6028,
2188
- "replacements": {}
2189
- },
2190
- {
2191
- "path": "training/turn-load/scripts/decide/headfile.py",
2192
- "source_sha256": "e6746d919b5aa9217d1e666b97dd09c807442c31d24fb04654dc22035a4014cd",
2193
- "sha256": "e6746d919b5aa9217d1e666b97dd09c807442c31d24fb04654dc22035a4014cd",
2194
- "source_bytes": 3414,
2195
- "bytes": 3414,
2196
- "replacements": {}
2197
- },
2198
- {
2199
- "path": "training/turn-load/scripts/decide/parity_probe.py",
2200
- "source_sha256": "ef4263b611d769752b0284d0ed17c538e0ae04a2652969b22f851584132a3d6f",
2201
- "sha256": "ef4263b611d769752b0284d0ed17c538e0ae04a2652969b22f851584132a3d6f",
2202
- "source_bytes": 6335,
2203
- "bytes": 6335,
2204
- "replacements": {}
2205
- },
2206
- {
2207
- "path": "training/turn-load/scripts/decide/rank-rubric.json",
2208
- "source_sha256": "f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98",
2209
- "sha256": "f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98",
2210
- "source_bytes": 139,
2211
- "bytes": 139,
2212
- "replacements": {}
2213
- },
2214
- {
2215
- "path": "training/turn-load/scripts/decide/replay_eval.py",
2216
- "source_sha256": "9fe36ccc8021cc62666630ac0f41ecf5c1d3a0b8a24ad438461f8120a9eb6228",
2217
- "sha256": "9fe36ccc8021cc62666630ac0f41ecf5c1d3a0b8a24ad438461f8120a9eb6228",
2218
- "source_bytes": 24237,
2219
- "bytes": 24237,
2220
- "replacements": {}
2221
- },
2222
- {
2223
- "path": "training/turn-load/scripts/decide/rerank/synthesize_sites.py",
2224
- "source_sha256": "60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d",
2225
- "sha256": "60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d",
2226
- "source_bytes": 15544,
2227
- "bytes": 15544,
2228
- "replacements": {}
2229
- },
2230
- {
2231
- "path": "training/turn-load/scripts/decide/rerank/train_rerank.py",
2232
- "source_sha256": "1fed73f53fd2e3512ec0d626edd3b41bb0969bf52ca302dd1fc4c772737f6e59",
2233
- "sha256": "1fed73f53fd2e3512ec0d626edd3b41bb0969bf52ca302dd1fc4c772737f6e59",
2234
- "source_bytes": 9686,
2235
- "bytes": 9686,
2236
- "replacements": {}
2237
- },
2238
- {
2239
- "path": "training/turn-load/scripts/decide/row.schema.json",
2240
- "source_sha256": "70d83cc6d22bd465cad54c696403be9bbfd7152460d74ee7ed9462046f377913",
2241
- "sha256": "70d83cc6d22bd465cad54c696403be9bbfd7152460d74ee7ed9462046f377913",
2242
- "source_bytes": 7261,
2243
- "bytes": 7261,
2244
- "replacements": {}
2245
- },
2246
- {
2247
- "path": "training/turn-load/scripts/decide/rows.py",
2248
- "source_sha256": "2554a171844b7a432920b82a0a9a7d070fac6e7777f171e9c53b8eed557a325e",
2249
- "sha256": "2554a171844b7a432920b82a0a9a7d070fac6e7777f171e9c53b8eed557a325e",
2250
- "source_bytes": 4127,
2251
- "bytes": 4127,
2252
- "replacements": {}
2253
- },
2254
- {
2255
- "path": "training/turn-load/scripts/decide/skill_discovery.py",
2256
- "source_sha256": "cb9bd0edd5472ba114310ba9f50a3d2089b2964696250ebea1c2ca9f0f0037e7",
2257
- "sha256": "cb9bd0edd5472ba114310ba9f50a3d2089b2964696250ebea1c2ca9f0f0037e7",
2258
- "source_bytes": 5956,
2259
- "bytes": 5956,
2260
- "replacements": {}
2261
- },
2262
- {
2263
- "path": "training/turn-load/scripts/decide/train.py",
2264
- "source_sha256": "18061510ed0299786f812f7b41d542ab4469d0d6148e9d93eb23bd7582d4b3b5",
2265
- "sha256": "18061510ed0299786f812f7b41d542ab4469d0d6148e9d93eb23bd7582d4b3b5",
2266
- "source_bytes": 30276,
2267
- "bytes": 30276,
2268
- "replacements": {}
2269
- },
2270
- {
2271
- "path": "training/unit-rank/mix.py",
2272
- "source_sha256": "0298ad3382a82c6138d48ce4d81aaa9384ebb755f775d802ff0ae7a8766b6685",
2273
- "sha256": "0298ad3382a82c6138d48ce4d81aaa9384ebb755f775d802ff0ae7a8766b6685",
2274
- "source_bytes": 824,
2275
- "bytes": 824,
2276
- "replacements": {}
2277
- },
2278
- {
2279
- "path": "training/unit-rank/prepare_rank_data.py",
2280
- "source_sha256": "9fea6d672870b24a829325d0c7d13ccceaf8927de35fe2af169b087d5653179b",
2281
- "sha256": "9fea6d672870b24a829325d0c7d13ccceaf8927de35fe2af169b087d5653179b",
2282
- "source_bytes": 2407,
2283
- "bytes": 2407,
2284
- "replacements": {}
2285
- },
2286
- {
2287
- "path": "training/unit-rank/requirements-cuda.txt",
2288
- "source_sha256": "4689a183766b25872744f0bc712e923560324572a25ab07a4a337104e7d80654",
2289
- "sha256": "4689a183766b25872744f0bc712e923560324572a25ab07a4a337104e7d80654",
2290
- "source_bytes": 123,
2291
- "bytes": 123,
2292
- "replacements": {}
2293
- },
2294
- {
2295
- "path": "training/turn-load/requirements-resolved.txt",
2296
- "source_sha256": "cd648d84da48acdc5312a3affd846a93bda39d7a6c94dbd1f446ec77b8872dd9",
2297
- "sha256": "cd648d84da48acdc5312a3affd846a93bda39d7a6c94dbd1f446ec77b8872dd9",
2298
- "source_bytes": 1040,
2299
- "bytes": 1040,
2300
- "replacements": {}
2301
- },
2302
- {
2303
- "path": "sessions/transcripts.jsonl.gz",
2304
- "source_sha256": "73189172526df6e48c297a073b9debff51d06e78b35ded75f96596715f78c537",
2305
- "sha256": "59b16f6b4d9668600ac499e30fc4f44e3ac58ca57dbf1d05f48388cf3d2d8369",
2306
- "rows": 3556,
2307
- "messages": 126133,
2308
- "replacements": {
2309
- "training_host": 78
2310
- }
2311
- },
2312
- {
2313
- "path": "training/unit-rank/scripts/artifact_paths.py",
2314
- "source_sha256": "2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54",
2315
- "sha256": "2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54",
2316
- "bytes": 3138,
2317
- "replacements": {}
2318
- },
2319
- {
2320
- "path": "training/turn-load/scripts/artifact_paths.py",
2321
- "source_sha256": "2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54",
2322
- "sha256": "2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54",
2323
- "bytes": 3138,
2324
- "replacements": {}
2325
- },
2326
- {
2327
- "path": "training/code-rank/scripts/artifact_paths.py",
2328
- "source_sha256": "2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54",
2329
- "sha256": "2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54",
2330
- "bytes": 3138,
2331
- "replacements": {}
2332
  }
2333
  ]
 
1
  [
2
  {
3
  "path": "train.jsonl",
4
+ "source_sha256": "ec3a0008eb903cd455912ce094a1e53884498a77ef76a20fa72805ab7b9aaeb5",
5
+ "sha256": "12c4b2d46ee1088882be996f409b661c2b9741c5e97cd6a71e3d580102633685",
6
+ "source_bytes": 27801342,
7
+ "bytes": 27820718,
8
+ "rows": 2746,
9
+ "relabel": "rows 2746, rows with a changed label 2557"
 
 
10
  },
11
  {
12
  "path": "val.jsonl",
13
+ "source_sha256": "a2f446c6b3b34b2dec1b70a04836ebc55e7951bcc4bcca0556019cb030dcda30",
14
+ "sha256": "06cc4a0225da9d167662942175efb63aa065d2da3ac5058ad9b6045096b8c051",
15
+ "source_bytes": 4998899,
16
+ "bytes": 5003330,
17
+ "rows": 489,
18
+ "relabel": "rows 489, rows with a changed label 445"
 
 
19
  },
20
  {
21
  "path": "holdout.jsonl",
22
+ "source_sha256": "f50ec8ba164e5b01536b02c5d62c81285dee3732822b1381e39a69b869ed80c9",
23
+ "sha256": "6ac4d873ce4802e10fc6178c4511508e36b88e86744335f0c5c6d9c23751fb4d",
24
+ "source_bytes": 7542702,
25
+ "bytes": 7548490,
26
+ "rows": 739,
27
+ "relabel": "rows 739, rows with a changed label 680"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
  }
29
  ]
NOTICE CHANGED
@@ -1,4 +1,4 @@
1
- Bialy training release
2
 
3
  Factory release material: Apache License 2.0; see LICENSE.
4
  Code excerpts and public test fixtures: original repository licenses and notices
 
1
+ Painted Wolf Decide training release
2
 
3
  Factory release material: Apache License 2.0; see LICENSE.
4
  Code excerpts and public test fixtures: original repository licenses and notices
PROVENANCE.json CHANGED
@@ -1,6 +1,9 @@
1
  {
2
  "schema": "pw-decide-training-release/1",
3
- "version": "open1-b5-e4",
 
 
 
4
  "sources": "Archived factory, training and evaluation artifacts; physical storage locations are private.",
5
  "models": [
6
  {
@@ -53,430 +56,5 @@
53
  "writer"
54
  ]
55
  }
56
- ],
57
- "repositories": [
58
- {
59
- "name": "cobra",
60
- "url": "https://github.com/spf13/cobra",
61
- "commit": "adbc8813901bba65827259daa8e22ff94ec1f30e",
62
- "spdx": "Apache-2.0",
63
- "language": "go",
64
- "split": "train",
65
- "license_files": {
66
- "LICENSE.txt": "5e3400b93bbb099e83e52bab885e7441750673c21f97988ca3f1240639b63283"
67
- }
68
- },
69
- {
70
- "name": "gin",
71
- "url": "https://github.com/gin-gonic/gin",
72
- "commit": "1bd0ecf2d3e08da37efa270bb7be20c1eddb0d68",
73
- "spdx": "MIT",
74
- "language": "go",
75
- "split": "train",
76
- "license_files": {
77
- "LICENSE": "b104efb2c7700691650f27034e8541c5ae0ed9af54d884287f19a7467ca2fe7f"
78
- }
79
- },
80
- {
81
- "name": "flask",
82
- "url": "https://github.com/pallets/flask",
83
- "commit": "d73fa1cdcbd8b1465c151db8924ba58b1dd14e35",
84
- "spdx": "BSD-3-Clause",
85
- "language": "python",
86
- "split": "train",
87
- "license_files": {
88
- "LICENSE.txt": "489a8e1108509ed98a37bb983e11e0f7e1d31f0bd8f99a79c8448e7ff37d07ea"
89
- }
90
- },
91
- {
92
- "name": "httpx",
93
- "url": "https://github.com/encode/httpx",
94
- "commit": "b5addb64f0161ff6bfe94c124ef76f6a1fba5254",
95
- "spdx": "BSD-3-Clause",
96
- "language": "python",
97
- "split": "train",
98
- "license_files": {
99
- "LICENSE.md": "4ec59d544f12b5f539a3a716fd321ac58ccd8030b465221f2c880200cdf28d8d"
100
- }
101
- },
102
- {
103
- "name": "requests",
104
- "url": "https://github.com/psf/requests",
105
- "commit": "611c6162cbc4ac2020a2f91c7cfa4f3abf9bbb60",
106
- "spdx": "Apache-2.0",
107
- "language": "python",
108
- "split": "train",
109
- "license_files": {
110
- "LICENSE": "09e8a9bcec8067104652c168685ab0931e7868f9c8284b66f5ae6edae5f1130b"
111
- }
112
- },
113
- {
114
- "name": "express",
115
- "url": "https://github.com/expressjs/express",
116
- "commit": "9a34acf03cb818ff3f8bc40e44176e277a25cbb9",
117
- "spdx": "MIT",
118
- "language": "javascript",
119
- "split": "train",
120
- "license_files": {
121
- "LICENSE": "95a5762890e5c1c9808921cef095661fc482c5e1f0bba31446ac85595df6237c"
122
- }
123
- },
124
- {
125
- "name": "axios",
126
- "url": "https://github.com/axios/axios",
127
- "commit": "2426e03ba9020be31ed013873423cea6b7cd2e67",
128
- "spdx": "MIT",
129
- "language": "javascript",
130
- "split": "train",
131
- "license_files": {
132
- "LICENSE": "82761059eaedacb3356803aea8a170d8298609f91b14fc32ee1bfb40d690183c"
133
- }
134
- },
135
- {
136
- "name": "gson",
137
- "url": "https://github.com/google/gson",
138
- "commit": "854c8255b625cf1e13c701a83ea9ccb4caaa576a",
139
- "spdx": "Apache-2.0",
140
- "language": "java",
141
- "split": "train",
142
- "license_files": {
143
- "LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
144
- }
145
- },
146
- {
147
- "name": "okhttp",
148
- "url": "https://github.com/square/okhttp",
149
- "commit": "40a3b8749deaacf60a04c890aed052cb14ad36ee",
150
- "spdx": "Apache-2.0",
151
- "language": "kotlin",
152
- "split": "train",
153
- "license_files": {
154
- "LICENSE.txt": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
155
- }
156
- },
157
- {
158
- "name": "axum",
159
- "url": "https://github.com/tokio-rs/axum",
160
- "commit": "bc8d4912435204a4a073cea70355ae0328d263fe",
161
- "spdx": "MIT",
162
- "language": "rust",
163
- "split": "train",
164
- "license_files": {
165
- "LICENSE": "008c87afcd2e626eaf564093250bed06dd7efb5732113264bba3dda8f1c556a1"
166
- }
167
- },
168
- {
169
- "name": "fd",
170
- "url": "https://github.com/sharkdp/fd",
171
- "commit": "ce97e473ebaec49697c07daa50a7bc2b32f713d2",
172
- "spdx": "MIT OR Apache-2.0",
173
- "language": "rust",
174
- "split": "train",
175
- "license_files": {
176
- "LICENSE-MIT": "322cfc7aa0c774d0eca3b2610f1d414de3ddbd7d8dd4b9dea941a13a6eb07455",
177
- "LICENSE-APACHE": "73c83c60d817e7df1943cb3f0af81e4939a8352c9a96c2fd00451b1116fa635c"
178
- }
179
- },
180
- {
181
- "name": "thor",
182
- "url": "https://github.com/rails/thor",
183
- "commit": "cb11e8e87b3cfe6bc9f6d4b092897f0dfda2476d",
184
- "spdx": "MIT",
185
- "language": "ruby",
186
- "split": "train",
187
- "license_files": {
188
- "LICENSE.md": "4df3e92a124b84d6767b1a0f1302600ea41d7616cda0b0fc40c1eefe4c9e414b"
189
- }
190
- },
191
- {
192
- "name": "slim",
193
- "url": "https://github.com/slimphp/Slim",
194
- "commit": "3675bf6baac66b07032575b7bef4200b60b7974b",
195
- "spdx": "MIT",
196
- "language": "php",
197
- "split": "train",
198
- "license_files": {
199
- "LICENSE.md": "24d84675cc9fffd507101fd43fecb5dd7f87679f065d90b2ef597d0a8b7e5994"
200
- }
201
- },
202
- {
203
- "name": "zod",
204
- "url": "https://github.com/colinhacks/zod",
205
- "commit": "2bf7b0630d5378033e90bcee82cb32b0fe04628e",
206
- "spdx": "MIT",
207
- "language": "typescript",
208
- "split": "holdout",
209
- "license_files": {
210
- "LICENSE": "3f1189b28e3866e0d979968d466b78f813f76827cfdca1fbb124cc0a5c8841f8"
211
- }
212
- },
213
- {
214
- "name": "sinatra",
215
- "url": "https://github.com/sinatra/sinatra",
216
- "commit": "cb22afd7902b566b6eaba6c4ea89739494a65d12",
217
- "spdx": "MIT",
218
- "language": "ruby",
219
- "split": "holdout",
220
- "license_files": {
221
- "LICENSE": "f2f60556a2d5c8d7ef9d54ec1c62152ff40d626b807aba5bdc5126f669ed500f"
222
- }
223
- },
224
- {
225
- "name": "ripgrep",
226
- "url": "https://github.com/BurntSushi/ripgrep",
227
- "commit": "3fce3b5bb0236da2df6d99672afb8a719642eca7",
228
- "spdx": "Unlicense OR MIT",
229
- "language": "rust",
230
- "split": "holdout",
231
- "license_files": {
232
- "LICENSE-MIT": "0f96a83840e146e43c0ec96a22ec1f392e0680e6c1226e6f3ba87e0740af850f",
233
- "COPYING": "01c266bced4a434da0051174d6bee16a4c82cf634e2679b6155d40d75012390f",
234
- "UNLICENSE": "7e12e5df4bae12cb21581ba157ced20e1986a0508dd10d0e8a4ab9a4cf94e85c"
235
- }
236
- }
237
- ],
238
- "code_rank": {
239
- "input_files": [
240
- "coderank/dumps/project_search-docs-axios.jsonl",
241
- "coderank/dumps/project_search-docs-axum.jsonl",
242
- "coderank/dumps/project_search-docs-cobra.jsonl",
243
- "coderank/dumps/project_search-docs-fd.jsonl",
244
- "coderank/dumps/project_search-docs-gin.jsonl",
245
- "coderank/dumps/project_search-docs-gson.jsonl",
246
- "coderank/dumps/project_search-docs-okhttp.jsonl",
247
- "coderank/dumps/project_search-docs-slim.jsonl",
248
- "coderank/dumps/project_search-docs-thor.jsonl",
249
- "coderank/dumps/project_search-model-axios.jsonl",
250
- "coderank/dumps/project_search-model-axum.jsonl",
251
- "coderank/dumps/project_search-model-cobra.jsonl",
252
- "coderank/dumps/project_search-model-express.jsonl",
253
- "coderank/dumps/project_search-model-fd.jsonl",
254
- "coderank/dumps/project_search-model-flask.jsonl",
255
- "coderank/dumps/project_search-model-gin.jsonl",
256
- "coderank/dumps/project_search-model-gson.jsonl",
257
- "coderank/dumps/project_search-model-httpx.jsonl",
258
- "coderank/dumps/project_search-model-okhttp.jsonl",
259
- "coderank/dumps/project_search-model-requests.jsonl",
260
- "coderank/dumps/project_search-model-slim.jsonl",
261
- "coderank/dumps/project_search-model-thor.jsonl",
262
- "coderank/dumps/repomap_tags-docs-axios.jsonl",
263
- "coderank/dumps/repomap_tags-docs-axum.jsonl",
264
- "coderank/dumps/repomap_tags-docs-cobra.jsonl",
265
- "coderank/dumps/repomap_tags-docs-fd.jsonl",
266
- "coderank/dumps/repomap_tags-docs-gin.jsonl",
267
- "coderank/dumps/repomap_tags-docs-gson.jsonl",
268
- "coderank/dumps/repomap_tags-docs-okhttp.jsonl",
269
- "coderank/dumps/repomap_tags-docs-slim.jsonl",
270
- "coderank/dumps/repomap_tags-docs-thor.jsonl",
271
- "coderank/dumps/repomap_tags-model-axios.jsonl",
272
- "coderank/dumps/repomap_tags-model-axum.jsonl",
273
- "coderank/dumps/repomap_tags-model-cobra.jsonl",
274
- "coderank/dumps/repomap_tags-model-express.jsonl",
275
- "coderank/dumps/repomap_tags-model-fd.jsonl",
276
- "coderank/dumps/repomap_tags-model-flask.jsonl",
277
- "coderank/dumps/repomap_tags-model-gin.jsonl",
278
- "coderank/dumps/repomap_tags-model-gson.jsonl",
279
- "coderank/dumps/repomap_tags-model-okhttp.jsonl",
280
- "coderank/dumps/repomap_tags-model-requests.jsonl",
281
- "coderank/dumps/repomap_tags-model-slim.jsonl",
282
- "coderank/dumps/repomap_tags-model-thor.jsonl",
283
- "coderank/dumps/summarize_definitions-docs-axios.jsonl",
284
- "coderank/dumps/summarize_definitions-docs-axum.jsonl",
285
- "coderank/dumps/summarize_definitions-docs-cobra.jsonl",
286
- "coderank/dumps/summarize_definitions-docs-fd.jsonl",
287
- "coderank/dumps/summarize_definitions-docs-gin.jsonl",
288
- "coderank/dumps/summarize_definitions-docs-gson.jsonl",
289
- "coderank/dumps/summarize_definitions-docs-httpx.jsonl",
290
- "coderank/dumps/summarize_definitions-docs-okhttp.jsonl",
291
- "coderank/dumps/summarize_definitions-docs-slim.jsonl",
292
- "coderank/dumps/summarize_definitions-docs-thor.jsonl",
293
- "coderank/dumps/summarize_definitions-model-axios.jsonl",
294
- "coderank/dumps/summarize_definitions-model-axum.jsonl",
295
- "coderank/dumps/summarize_definitions-model-cobra.jsonl",
296
- "coderank/dumps/summarize_definitions-model-express.jsonl",
297
- "coderank/dumps/summarize_definitions-model-fd.jsonl",
298
- "coderank/dumps/summarize_definitions-model-flask.jsonl",
299
- "coderank/dumps/summarize_definitions-model-gin.jsonl",
300
- "coderank/dumps/summarize_definitions-model-gson.jsonl",
301
- "coderank/dumps/summarize_definitions-model-httpx.jsonl",
302
- "coderank/dumps/summarize_definitions-model-okhttp.jsonl",
303
- "coderank/dumps/summarize_definitions-model-requests.jsonl",
304
- "coderank/dumps/summarize_definitions-model-slim.jsonl",
305
- "coderank/dumps/summarize_definitions-model-thor.jsonl",
306
- "coderank/dumps/summarize_structure-docs-axios.jsonl",
307
- "coderank/dumps/summarize_structure-docs-axum.jsonl",
308
- "coderank/dumps/summarize_structure-docs-cobra.jsonl",
309
- "coderank/dumps/summarize_structure-docs-fd.jsonl",
310
- "coderank/dumps/summarize_structure-docs-gin.jsonl",
311
- "coderank/dumps/summarize_structure-docs-gson.jsonl",
312
- "coderank/dumps/summarize_structure-docs-httpx.jsonl",
313
- "coderank/dumps/summarize_structure-docs-okhttp.jsonl",
314
- "coderank/dumps/summarize_structure-docs-slim.jsonl",
315
- "coderank/dumps/summarize_structure-docs-thor.jsonl",
316
- "coderank/dumps/summarize_structure-model-axios.jsonl",
317
- "coderank/dumps/summarize_structure-model-axum.jsonl",
318
- "coderank/dumps/summarize_structure-model-cobra.jsonl",
319
- "coderank/dumps/summarize_structure-model-express.jsonl",
320
- "coderank/dumps/summarize_structure-model-fd.jsonl",
321
- "coderank/dumps/summarize_structure-model-flask.jsonl",
322
- "coderank/dumps/summarize_structure-model-gin.jsonl",
323
- "coderank/dumps/summarize_structure-model-gson.jsonl",
324
- "coderank/dumps/summarize_structure-model-httpx.jsonl",
325
- "coderank/dumps/summarize_structure-model-okhttp.jsonl",
326
- "coderank/dumps/summarize_structure-model-requests.jsonl",
327
- "coderank/dumps/summarize_structure-model-slim.jsonl",
328
- "coderank/dumps/summarize_structure-model-thor.jsonl",
329
- "coderank/sites/rows-sites-axios.jsonl",
330
- "coderank/sites/rows-sites-axum.jsonl",
331
- "coderank/sites/rows-sites-cobra.jsonl",
332
- "coderank/sites/rows-sites-express.jsonl",
333
- "coderank/sites/rows-sites-fd.jsonl",
334
- "coderank/sites/rows-sites-flask.jsonl",
335
- "coderank/sites/rows-sites-gin.jsonl",
336
- "coderank/sites/rows-sites-gson.jsonl",
337
- "coderank/sites/rows-sites-httpx.jsonl",
338
- "coderank/sites/rows-sites-okhttp.jsonl",
339
- "coderank/sites/rows-sites-requests.jsonl",
340
- "coderank/sites/rows-sites-slim.jsonl",
341
- "coderank/sites/rows-sites-thor.jsonl"
342
- ],
343
- "input_rows": 44671,
344
- "training_rows": 40203,
345
- "validation_rows": 4468,
346
- "training_examples": 251360,
347
- "validation_examples": 27828,
348
- "seed": 7,
349
- "epochs": 6,
350
- "batch_size": 64,
351
- "max_length": 512,
352
- "learning_rate": 0.0003,
353
- "validation_fraction": 0.1,
354
- "excluded_repository_names": [
355
- "zod",
356
- "sinatra",
357
- "ripgrep"
358
- ],
359
- "split_method": "shuffle rows in listed file order with Python random.Random(7), then first int(n * 0.9) rows for training; continue the same RNG for candidate sampling"
360
- },
361
- "sessions": {
362
- "sessions": 3556,
363
- "messages": 126133,
364
- "database_shards": 385,
365
- "coverage": {
366
- "train.jsonl": {
367
- "sessions": 2222,
368
- "sessions_with_transcripts": 2222,
369
- "sessions_without_transcripts": 0
370
- },
371
- "val.jsonl": {
372
- "sessions": 399,
373
- "sessions_with_transcripts": 399,
374
- "sessions_without_transcripts": 0
375
- },
376
- "holdout.jsonl": {
377
- "sessions": 595,
378
- "sessions_with_transcripts": 595,
379
- "sessions_without_transcripts": 0
380
- }
381
- },
382
- "snapshot_rows": 3741,
383
- "later_rows": 233,
384
- "historical_judged_rows": 3974,
385
- "export_scope": "All sessions and selected conversation fields from 385 archived pass1v3b shard databases. No WAL files were present. Raw databases, attachments, spill files, provider settings and unrelated host state are not included."
386
- },
387
- "training": {
388
- "turn_load": {
389
- "label": "open1-turn-load-B5-release-independent",
390
- "train": "train.jsonl",
391
- "validation": "val.jsonl",
392
- "corpus": "corpora/independent.json",
393
- "trainer_commit": "3d79a91c928cbafd254e37de1798b75b8c8df9b1",
394
- "recipe": {
395
- "families": "tools",
396
- "tool_truth": "consensus",
397
- "tool_encoding": "independent",
398
- "tool_negatives": 24,
399
- "inverse_negative_sampling_weights": true,
400
- "tool_weight": "none",
401
- "positive_weight": 6,
402
- "seed": 11,
403
- "batch_size": 64,
404
- "learning_rate": 0.0005,
405
- "maximum_epochs": 45,
406
- "patience": 8,
407
- "context_tokens": 1024,
408
- "head_tokens": 512,
409
- "tool_option_words": 60
410
- }
411
- },
412
- "unit_rank": {
413
- "label": "open1-unit-rank-e4-dense1",
414
- "train": "unit-rank/train.jsonl",
415
- "selection": "unit-rank/selection.jsonl",
416
- "corpus": "corpus.json",
417
- "trainer_commit": "946aea8506c9440a662f87a8d5479963f27735d4",
418
- "factory_commit": "b60c15122b4dcddaee75bc45b2e9413a0fe3fa8e",
419
- "recipe": {
420
- "families": "skills,requests",
421
- "rank_levels": "skills-blended",
422
- "skill_scored": 8,
423
- "skill_zeros": 12,
424
- "seed": 11,
425
- "batch_size": 32,
426
- "mixed_generated_family_fraction": 0.5,
427
- "mix_seed": 7
428
- }
429
- },
430
- "code_rank": {
431
- "label": "open1-code-rank",
432
- "trainer_revision": "21e0e2d108",
433
- "recipe": "training/code-rank/recipe.json"
434
- }
435
- },
436
- "privacy": {
437
- "known_infrastructure_replacements": {
438
- "archive_path": 824,
439
- "personal_path": 0,
440
- "training_host": 78,
441
- "private_host": 0,
442
- "private_identity": 0,
443
- "credential": 0
444
- },
445
- "policy": "Known infrastructure references replaced; training labels and row ordering preserved. Public upstream code/test-fixture strings remain under their original licenses. Original byte hashes and release byte hashes are listed in FILES.json."
446
- },
447
- "limitations": [
448
- "B5 evidence is validation and selection, including cutoff 0.94; not fresh end-to-end acceptance.",
449
- "E4 historical acceptance failed the common-skill criterion. Other listed skill criteria passed.",
450
- "Historical acceptance used an earlier policy; later host tests establish scoped implementation correctness, not model task quality.",
451
- "Historical agreement.jsonl belongs to original open1 judging, not the later B5/E4 labels.",
452
- "The generic audit dataset command assumes a single uniform corpus and split and does not establish validity of this multi-configuration release.",
453
- "Model regeneration has not been run; exact weights may depend on runtime, dependencies, hardware and model availability.",
454
- "Auxiliary deployment logs and raw shard state are outside this public bundle."
455
- ],
456
- "selected_heads": [
457
- {
458
- "file": "turn-load.safetensors",
459
- "sha256": "550f30b94d5c6a8680d2cb2a607ebc9ee33e709c7679f72606d1e9fda17f2e90",
460
- "bytes": 59881652
461
- },
462
- {
463
- "file": "unit-rank.safetensors",
464
- "sha256": "c62674993eabba94660eccfafbb6e04d1549adc7335656c0ddefa7f068fa58f7",
465
- "bytes": 59881268
466
- },
467
- {
468
- "file": "code-rank.safetensors",
469
- "sha256": "abe235c347a223e957a0f8b7b979e3a584aebab76ee239e5f62117e839b9c40a",
470
- "bytes": 59880964
471
- }
472
- ],
473
- "backbone": {
474
- "model": "convaiinnovations/laya-multilingual",
475
- "revision": "e4e9ddf21a7b1903b7acffd8814ad4307bf63a67"
476
- },
477
- "public_name": "Bialy",
478
- "package_version": "open1-b5-e4-bialy",
479
- "model_repo": "paintedwolfcode/bialy",
480
- "code_repo": "https://github.com/paintedwolf-ai/bialy",
481
- "dataset_repo": "paintedwolfcode/bialy-dataset"
482
  }
 
1
  {
2
  "schema": "pw-decide-training-release/1",
3
+ "version": "open1-b7g-e4",
4
+ "derived_from": "open1-b5-e4",
5
+ "derived_on": "2026-09-30",
6
+ "derivation": "labels.guides relabelled from tool calls: a unit is needed when its turn called a tool it attaches to or one it declares needed_with (train-host/relabel_guides.py); rows, order, and other labels unchanged",
7
  "sources": "Archived factory, training and evaluation artifacts; physical storage locations are private.",
8
  "models": [
9
  {
 
56
  "writer"
57
  ]
58
  }
59
+ ]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
  }
README.md CHANGED
@@ -1,16 +1,6 @@
1
  ---
2
  license: apache-2.0
3
- pretty_name: Bialy open1 B5 E4 training release
4
- language:
5
- - en
6
- - de
7
- - fr
8
- - es
9
- - pt
10
- - it
11
- - ja
12
- - zh
13
- - ko
14
  tags:
15
  - synthetic
16
  - coding-agents
@@ -24,230 +14,20 @@ configs:
24
  path: val.jsonl
25
  - split: test
26
  path: holdout.jsonl
27
- - config_name: unit_rank
28
- data_files:
29
- - split: train
30
- path: unit-rank/train.jsonl
31
- - split: validation
32
- path: unit-rank/selection.jsonl
33
- - split: test
34
- path: unit-rank/historical-acceptance.jsonl
35
- - config_name: code_rank_inputs
36
- data_files:
37
- - split: unsplit
38
- path:
39
- - coderank/dumps/project_search-docs-axios.jsonl
40
- - coderank/dumps/project_search-docs-axum.jsonl
41
- - coderank/dumps/project_search-docs-cobra.jsonl
42
- - coderank/dumps/project_search-docs-fd.jsonl
43
- - coderank/dumps/project_search-docs-gin.jsonl
44
- - coderank/dumps/project_search-docs-gson.jsonl
45
- - coderank/dumps/project_search-docs-okhttp.jsonl
46
- - coderank/dumps/project_search-docs-slim.jsonl
47
- - coderank/dumps/project_search-docs-thor.jsonl
48
- - coderank/dumps/project_search-model-axios.jsonl
49
- - coderank/dumps/project_search-model-axum.jsonl
50
- - coderank/dumps/project_search-model-cobra.jsonl
51
- - coderank/dumps/project_search-model-express.jsonl
52
- - coderank/dumps/project_search-model-fd.jsonl
53
- - coderank/dumps/project_search-model-flask.jsonl
54
- - coderank/dumps/project_search-model-gin.jsonl
55
- - coderank/dumps/project_search-model-gson.jsonl
56
- - coderank/dumps/project_search-model-httpx.jsonl
57
- - coderank/dumps/project_search-model-okhttp.jsonl
58
- - coderank/dumps/project_search-model-requests.jsonl
59
- - coderank/dumps/project_search-model-slim.jsonl
60
- - coderank/dumps/project_search-model-thor.jsonl
61
- - coderank/dumps/repomap_tags-docs-axios.jsonl
62
- - coderank/dumps/repomap_tags-docs-axum.jsonl
63
- - coderank/dumps/repomap_tags-docs-cobra.jsonl
64
- - coderank/dumps/repomap_tags-docs-fd.jsonl
65
- - coderank/dumps/repomap_tags-docs-gin.jsonl
66
- - coderank/dumps/repomap_tags-docs-gson.jsonl
67
- - coderank/dumps/repomap_tags-docs-okhttp.jsonl
68
- - coderank/dumps/repomap_tags-docs-slim.jsonl
69
- - coderank/dumps/repomap_tags-docs-thor.jsonl
70
- - coderank/dumps/repomap_tags-model-axios.jsonl
71
- - coderank/dumps/repomap_tags-model-axum.jsonl
72
- - coderank/dumps/repomap_tags-model-cobra.jsonl
73
- - coderank/dumps/repomap_tags-model-express.jsonl
74
- - coderank/dumps/repomap_tags-model-fd.jsonl
75
- - coderank/dumps/repomap_tags-model-flask.jsonl
76
- - coderank/dumps/repomap_tags-model-gin.jsonl
77
- - coderank/dumps/repomap_tags-model-gson.jsonl
78
- - coderank/dumps/repomap_tags-model-okhttp.jsonl
79
- - coderank/dumps/repomap_tags-model-requests.jsonl
80
- - coderank/dumps/repomap_tags-model-slim.jsonl
81
- - coderank/dumps/repomap_tags-model-thor.jsonl
82
- - coderank/dumps/summarize_definitions-docs-axios.jsonl
83
- - coderank/dumps/summarize_definitions-docs-axum.jsonl
84
- - coderank/dumps/summarize_definitions-docs-cobra.jsonl
85
- - coderank/dumps/summarize_definitions-docs-fd.jsonl
86
- - coderank/dumps/summarize_definitions-docs-gin.jsonl
87
- - coderank/dumps/summarize_definitions-docs-gson.jsonl
88
- - coderank/dumps/summarize_definitions-docs-httpx.jsonl
89
- - coderank/dumps/summarize_definitions-docs-okhttp.jsonl
90
- - coderank/dumps/summarize_definitions-docs-slim.jsonl
91
- - coderank/dumps/summarize_definitions-docs-thor.jsonl
92
- - coderank/dumps/summarize_definitions-model-axios.jsonl
93
- - coderank/dumps/summarize_definitions-model-axum.jsonl
94
- - coderank/dumps/summarize_definitions-model-cobra.jsonl
95
- - coderank/dumps/summarize_definitions-model-express.jsonl
96
- - coderank/dumps/summarize_definitions-model-fd.jsonl
97
- - coderank/dumps/summarize_definitions-model-flask.jsonl
98
- - coderank/dumps/summarize_definitions-model-gin.jsonl
99
- - coderank/dumps/summarize_definitions-model-gson.jsonl
100
- - coderank/dumps/summarize_definitions-model-httpx.jsonl
101
- - coderank/dumps/summarize_definitions-model-okhttp.jsonl
102
- - coderank/dumps/summarize_definitions-model-requests.jsonl
103
- - coderank/dumps/summarize_definitions-model-slim.jsonl
104
- - coderank/dumps/summarize_definitions-model-thor.jsonl
105
- - coderank/dumps/summarize_structure-docs-axios.jsonl
106
- - coderank/dumps/summarize_structure-docs-axum.jsonl
107
- - coderank/dumps/summarize_structure-docs-cobra.jsonl
108
- - coderank/dumps/summarize_structure-docs-fd.jsonl
109
- - coderank/dumps/summarize_structure-docs-gin.jsonl
110
- - coderank/dumps/summarize_structure-docs-gson.jsonl
111
- - coderank/dumps/summarize_structure-docs-httpx.jsonl
112
- - coderank/dumps/summarize_structure-docs-okhttp.jsonl
113
- - coderank/dumps/summarize_structure-docs-slim.jsonl
114
- - coderank/dumps/summarize_structure-docs-thor.jsonl
115
- - coderank/dumps/summarize_structure-model-axios.jsonl
116
- - coderank/dumps/summarize_structure-model-axum.jsonl
117
- - coderank/dumps/summarize_structure-model-cobra.jsonl
118
- - coderank/dumps/summarize_structure-model-express.jsonl
119
- - coderank/dumps/summarize_structure-model-fd.jsonl
120
- - coderank/dumps/summarize_structure-model-flask.jsonl
121
- - coderank/dumps/summarize_structure-model-gin.jsonl
122
- - coderank/dumps/summarize_structure-model-gson.jsonl
123
- - coderank/dumps/summarize_structure-model-httpx.jsonl
124
- - coderank/dumps/summarize_structure-model-okhttp.jsonl
125
- - coderank/dumps/summarize_structure-model-requests.jsonl
126
- - coderank/dumps/summarize_structure-model-slim.jsonl
127
- - coderank/dumps/summarize_structure-model-thor.jsonl
128
- - coderank/sites/rows-sites-axios.jsonl
129
- - coderank/sites/rows-sites-axum.jsonl
130
- - coderank/sites/rows-sites-cobra.jsonl
131
- - coderank/sites/rows-sites-express.jsonl
132
- - coderank/sites/rows-sites-fd.jsonl
133
- - coderank/sites/rows-sites-flask.jsonl
134
- - coderank/sites/rows-sites-gin.jsonl
135
- - coderank/sites/rows-sites-gson.jsonl
136
- - coderank/sites/rows-sites-httpx.jsonl
137
- - coderank/sites/rows-sites-okhttp.jsonl
138
- - coderank/sites/rows-sites-requests.jsonl
139
- - coderank/sites/rows-sites-slim.jsonl
140
- - coderank/sites/rows-sites-thor.jsonl
141
- - config_name: sessions
142
- data_files:
143
- - split: archive
144
- path: sessions/transcripts.jsonl.gz
145
- - config_name: tasks
146
- data_files:
147
- - split: archive
148
- path: tasks.jsonl
149
  ---
150
 
151
- # Bialy training release open1-b5-e4
152
-
153
- Package revision `open1-b5-e4-bialy` updates public branding only. The historical
154
- model release identity, training data and trainer snapshots are unchanged.
155
-
156
- Archived synthetic training inputs and conversation exports for the B5 turn-load,
157
- E4 unit-rank, and open1 code-rank heads. This is the companion dataset for
158
- `paintedwolfcode/bialy`. Preparation does not publish either repository.
159
- The model weights are distributed separately, unchanged.
160
 
161
- ## Contents and lineage
 
 
 
162
 
163
  | Configuration | Files | Meaning |
164
  | --- | --- | --- |
165
- | B5 turn-load | `train.jsonl` (2,746), `val.jsonl` (489), `corpora/independent.json` | Exact archived row order; infrastructure metadata sanitized where present. B5 uses independent tool scoring and up to 60-word options. |
166
- | E4 unit-rank | `unit-rank/train.jsonl` (3,792), `unit-rank/selection.jsonl` (1,024), `corpus.json` | Archived mixed training input and selection set. The mixture selects half of generated families with seed 7 and uses the session rows. |
167
- | Generated skill requests | `unit-rank/generated-train.jsonl` (2,634) | Input pool for the E4 mixture; do not append it to the already mixed training file. |
168
- | Historical acceptance | `unit-rank/historical-acceptance.jsonl` (789) | E4's historical synthetic acceptance requests; not current-policy acceptance. |
169
- | Session holdout | `holdout.jsonl` (739) | Archived held-out repository rows, not a fresh end-to-end evaluation. |
170
- | Original open1 labels | `historical/`, `corpora/original-open1.json`, `agreement.jsonl` | Original judging and splits before later B5/E4 judging. Agreement belongs to these original labels. |
171
- | Code-rank | `coderank/dumps/`, `coderank/sites/` | Actual candidate inputs. `training/code-rank/recipe.json` lists the selected nonempty files in training order. |
172
- | Code generation inputs | `coderank/pairs/`, `coderank/units/` | Generated requests and harvested code units, including held-out repositories. These are not all training inputs. |
173
- | Driven tasks | `tasks.jsonl` (3,024) | Generated prompts, follow-ups, models and recorded outcomes. |
174
- | Conversation exports | `sessions/transcripts.jsonl.gz` | 3,556 sessions and 126,133 messages from 385 archived shard databases, including child sessions. |
175
-
176
- The code-rank script excludes zod, sinatra, and ripgrep files, shuffles the selected
177
- rows with seed 7, then takes 90% for training before candidate sampling. Its
178
- `unsplit` viewer configuration is the input pool, not a pre-separated training set.
179
- The extracted example count is 251,360, matching the selected head metadata.
180
- All 116 archived dump files and 13 site files are included; empty and held-out
181
- files remain available but are excluded from the recorded training selection.
182
-
183
- The original session collection merged finished and live snapshots by
184
- `(root_session, session, receipt)`. `sessions/combined-snapshot-rows.jsonl`
185
- contains 3,741 rows; `sessions/later-rows.jsonl` adds 233 distinct rows. Their
186
- union is the 3,974-row historical judged set. The other snapshot exports overlap;
187
- do not concatenate all files under `sessions/` or combine configurations as one
188
- training split. Recovery metadata and later judging explain why the current
189
- training rows are not identical to the original labels.
190
-
191
- ## What is reproducible
192
-
193
- `FILES.json` records source and released byte hashes, row counts and infrastructure
194
- replacement counts. `SHA256SUMS` covers the complete release. The source originals
195
- remain privately archived; hashes do not imply that private originals are public.
196
- Training labels and row order are retained. Machine references are replaced with
197
- explicit placeholders; those changes are reflected in the released hashes.
198
-
199
- The two tool corpora intentionally differ: joint B2 uses six-word tool options,
200
- while independent B5 uses up to 60 words. The source corpus hashes are
201
- `51e3a65258fc34581bc8cfaae0f7c97f444f0a50f3dda906c77b6554fd5c66d3`
202
- and `c77abd8b40965c05cc43a7545e8c2db2405bdcfc18a73bad254632d523b67bcd`.
203
- The source validation rows hash is
204
- `67c4c86e3385f0241048f2633d97a4aad78a53a3b7a540592937a20eba669fd3`.
205
- Compare the two complete configurations, including their encoding differences;
206
- do not require equal corpus hashes or substitute one corpus for the other.
207
-
208
- `training/` includes historical trainer source snapshots and recipes. See
209
- [Training](TRAINING.md) for paths and arguments. Re-training was not run during
210
- release preparation. Identical model bytes are not promised across hardware,
211
- library versions, model availability, or nondeterministic GPU operations.
212
- Archived commit IDs identify history; they do not promise those commits exist in
213
- a subsequently flattened Git repository.
214
-
215
- The conversation export contains session relationships and ordered message fields,
216
- including structured content, tool calls/results, reasoning, and compaction fields
217
- where stored. It excludes raw databases, provider configuration, identities of
218
- machine operators, attachments and spill-file payloads. It is an export of retained
219
- messages, not a claim that every historical runtime event or pre-compaction token
220
- survived. `PROVENANCE.json` reports training-row session coverage.
221
-
222
- ## Evidence boundaries
223
-
224
- B5 results, including the 0.94 cutoff, are validation/selection evidence. E4's
225
- historical acceptance failed the common-skill criterion; its other listed skill
226
- criteria passed. Neither establishes fresh end-to-end acceptance under current
227
- host policy. Later host tests and startup checks establish implementation
228
- correctness within their scope, not model task quality. No new model work or
229
- acceptance evaluation was performed for this dataset.
230
-
231
- The factory's generic `audit dataset` assumes one uniform split/corpus and does
232
- not establish validity of this multi-configuration historical release. Use the
233
- checksum anchor and per-file provenance to verify artifact identity. Do not
234
- present a checksum check as a model-quality audit.
235
-
236
- ## Attribution and licenses
237
-
238
- Factory-authored release material is Apache-2.0. Repository code, comments, test
239
- fixtures and excerpts in the candidate inputs and conversations retain their
240
- upstream licenses; this bundle does not relicense them. `licenses/` contains the
241
- license texts at the archived pinned commits, with hashes checked against the
242
- repository manifest, and the applicable Requests notice. Public upstream test
243
- fixtures can contain example key material and example filesystem paths.
244
-
245
- `PROVENANCE.json` lists repository revisions and the models recorded by the
246
- factory. Original sessions and code requests used Qwen and Gemma; the original
247
- judging and later hosted judging used different configurations, retained under
248
- `evidence/`. The later judge/writer configuration records GLM and Inkling.
249
- Model identities and license reviews are historical provenance records.
250
 
251
- The decision heads build on [Laya](https://github.com/NandhaKishorM/laya), by
252
- Convai Innovations, and the multilingual Laya checkpoint over mmBERT-base.
253
- The upstream model weights are not included in this dataset.
 
1
  ---
2
  license: apache-2.0
3
+ pretty_name: Painted Wolf Decide training release open1-b7g-e4
 
 
 
 
 
 
 
 
 
 
4
  tags:
5
  - synthetic
6
  - coding-agents
 
14
  path: val.jsonl
15
  - split: test
16
  path: holdout.jsonl
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
17
  ---
18
 
19
+ # Painted Wolf Decide training release open1-b7g-e4
 
 
 
 
 
 
 
 
20
 
21
+ The turn-load rows of open1-b5-e4 with their guide-unit labels relabelled from tool calls,
22
+ the companion dataset for the guide-load head. Rows, order, sanitisation, and every other
23
+ label are those of the parent release; open1-b5-e4 remains the source for the unit-rank and
24
+ code-rank inputs and the conversation exports.
25
 
26
  | Configuration | Files | Meaning |
27
  | --- | --- | --- |
28
+ | Turn-load | train.jsonl (2746), val.jsonl (489), holdout.jsonl (739), `corpora/independent.json` | Parent rows; `labels.guides` is whether the turn called a tool the unit attaches to or is needed with. |
29
+ | Corpus | `corpus.json` | The host's unit, tool, and skill catalog, with each unit's `attaches` and `needed_with`. |
30
+ | Trainer | `training/turn-load/` | The trainer snapshot, recipes, relabelling, probe, and audit scripts. |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
 
32
+ `FILES.json` records the parent and released row hashes and the relabelling counts;
33
+ `SHA256SUMS` covers the release. See [Training](TRAINING.md) and `PROVENANCE.json`.
 
SHA256SUMS CHANGED
@@ -1,275 +1,34 @@
1
- f714ff4bf4880da24e0d64203287fce424ba7ff8392bd7b4abf1eb922b338d7f FILES.json
2
  3c2307bbbbb88f79560a165b098c600da8f20fb1ec024586d4be192ebc9c6f8e LICENSE
3
- 6ed8167d9fad68eb2952c1ab8fe35952173e770196251986c0dc868f042a74af NOTICE
4
- b31dd8ae8d0e3913b423815fc210f3fa4d15604684c0682a2f51a66ba65b04c8 PROVENANCE.json
5
- 96ddf668cc3280f3e8b37be0e5779481abe8e7f9c812976e1a93ce7cabe31a6b README.md
6
- cfda62952c0723efa27abba840acbc8848b500339aaf3cee68e41a5fe556fafe TRAINING.md
7
  7e09a014dd94f32d95196a3531e046c1904d526a593b670f448f1d87168a44ff agreement.jsonl
8
- 5bcfcc5cf524bf153544e076f8ee04377dd328a087af6655a9ac86c95ec29559 coderank/dumps/project_search-docs-axios.jsonl
9
- eeb63cd18accad559cb170cb32c60909acbea46105ed77d3a5c0db74ca190f3a coderank/dumps/project_search-docs-axum.jsonl
10
- 883f73dc55ac6fd11419cda77b6ff87a7ca6c58fdb0dcb6956b4e004cf44098e coderank/dumps/project_search-docs-cobra.jsonl
11
- 79b5695acd315bb10c178c51d7c717014d5fcb3a9daeaad780bde438a646e8ee coderank/dumps/project_search-docs-fd.jsonl
12
- c605944583ec4cb997a63d1db919d19d51958deec276325aa798f737d0aee35c coderank/dumps/project_search-docs-gin.jsonl
13
- 810915dab0f649e75d437b4c2399ca9b46ca71884d23f926c43ef205a5c53d57 coderank/dumps/project_search-docs-gson.jsonl
14
- e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 coderank/dumps/project_search-docs-httpx.jsonl
15
- 509e68398d0e8b22b3b60232132e84f671f2ddcc556a41aa167dc2888f274c72 coderank/dumps/project_search-docs-okhttp.jsonl
16
- 3375be55202398c3bac09c742659540fad16c2905def71335990b3d146b970ab coderank/dumps/project_search-docs-ripgrep.jsonl
17
- 411077eb35f38a817a11b4373209967a1cd6364149b89d55d43806bdd476e1cf coderank/dumps/project_search-docs-sinatra.jsonl
18
- 4d83ae26b127a67ba643260ba431a3d1ca0f3b84261fe2f7cedb92ada74c7312 coderank/dumps/project_search-docs-slim.jsonl
19
- 7e5bf295be859366377883ffdced6715f952b1f743f95eb10d7bf66fc278daac coderank/dumps/project_search-docs-thor.jsonl
20
- cf7b8e1cfc6db1be72ca542c1073e556121046b92c93a1b5646f1b7816060a64 coderank/dumps/project_search-docs-zod.jsonl
21
- e8665c1b219ba19ed8b9efc04ec9a59abbd5b3e5ac7245244ab46e0907fb3773 coderank/dumps/project_search-model-axios.jsonl
22
- e396c406a99d6b0b2adc663b07b6902aa1dba2738c0990fb6ee5cba8c7eb8b71 coderank/dumps/project_search-model-axum.jsonl
23
- f758dd450ffde967901524f830bb82809b23b74676d7dfd8192ef90f2525a109 coderank/dumps/project_search-model-cobra.jsonl
24
- 4e10be53a860d918c974c6f7c0df495b2057ba333ba32f6d5f27a7be9eb6d07b coderank/dumps/project_search-model-express.jsonl
25
- 8e49366dfbdb60e101178ad8176dcc457debd759037881dec2b34c8cf0ff76d1 coderank/dumps/project_search-model-fd.jsonl
26
- 4f296ad85a8098aa8157d1c961594a37a2e161e0243532d33a44697d144a44bd coderank/dumps/project_search-model-flask.jsonl
27
- d2dce8f7c3ebe84d49bc0be4bacc963c91534c8e667ac465c195c329e25ef447 coderank/dumps/project_search-model-gin.jsonl
28
- 624c9436208e17e12e5455640e9247a6dd352746093b54baf1a297bc33766893 coderank/dumps/project_search-model-gson.jsonl
29
- 5f3bc5225a03041c130090e21a92cf4178410aed4e1d13f04c7c3b3d5a1cc597 coderank/dumps/project_search-model-httpx.jsonl
30
- ab119fcc6ec66db1962cc362fad8657ee7c78fc409c1de8923b36fb8261e58b6 coderank/dumps/project_search-model-okhttp.jsonl
31
- 150474fd41d416e62c33c2db20573117d4b2dc1e4ee0c06646e396554f83984e coderank/dumps/project_search-model-requests.jsonl
32
- c099cfc6a427392781529bd41485d80d54b0ecd5b73eb1a32877f629f7f25b65 coderank/dumps/project_search-model-ripgrep.jsonl
33
- b76fb56eba15cb615fc7ae548f91928dc3be4df71d051248c48171529cb0b699 coderank/dumps/project_search-model-sinatra.jsonl
34
- a7836761487527798a4d8298cfcc7eb2582760658a6e09d2365cc56c230fe87e coderank/dumps/project_search-model-slim.jsonl
35
- 0b84460ad6c4e8cac36686add70716b975cdb1c6c2f8059d818eb8e28afb2222 coderank/dumps/project_search-model-thor.jsonl
36
- 580632b2fa373fd9960ddcd85c42aa8aaa6d2c3e76d9b6219af67098b4e15e2e coderank/dumps/project_search-model-zod.jsonl
37
- f695f184413b1cbcae84bff755f39d46873cae49064f6fa3ed17f1cfa2c9a9b4 coderank/dumps/repomap_tags-docs-axios.jsonl
38
- 3ceb783a0aa1379cd57278686978e013e61beb7d284e3bfc0512120366c2883e coderank/dumps/repomap_tags-docs-axum.jsonl
39
- 57149dc1d9a137a8f571e06e11e06f5df8a9aade5faf3ca6ade8a98733529807 coderank/dumps/repomap_tags-docs-cobra.jsonl
40
- d27ff9f47ef9045f82116151ce938dcca0da3b0329c74bafa34c7ed1ed0310aa coderank/dumps/repomap_tags-docs-fd.jsonl
41
- d58baa5d84cd9443c199ef40d01b46dda3e63f9cc225ccdd2d0075e96777a9f8 coderank/dumps/repomap_tags-docs-gin.jsonl
42
- 489c03a4cf47fab381c57c60afb15c773e704f8794e97d128c826d0b4b6ce059 coderank/dumps/repomap_tags-docs-gson.jsonl
43
- e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 coderank/dumps/repomap_tags-docs-httpx.jsonl
44
- 6f60069dbcfd26a2e63bfbed00f18609926c3178a80600a0dc05bb7d737688b7 coderank/dumps/repomap_tags-docs-okhttp.jsonl
45
- 70e1458de74c9c0a31d0351de01cdab23d435ae28b9ac383d68459c58458e30b coderank/dumps/repomap_tags-docs-ripgrep.jsonl
46
- e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 coderank/dumps/repomap_tags-docs-sinatra.jsonl
47
- e4ee8169ee1d5b267e5a3243551a5806a787e156470418940fbbb88c17200f48 coderank/dumps/repomap_tags-docs-slim.jsonl
48
- 673053bab8ebff64080b7a45a86827da139f245f230cd11814ec8a6437d3f724 coderank/dumps/repomap_tags-docs-thor.jsonl
49
- 259bd08ef58038d41f36e3d2c64110417e9891d761ef8375121f0e3541e14eea coderank/dumps/repomap_tags-docs-zod.jsonl
50
- 229737da35f6ebee09806e031df8d5ac2030208c0786e148d2e7129ea1fff941 coderank/dumps/repomap_tags-model-axios.jsonl
51
- 7d4a8f7d2e23d2d5b9d62713c290ad1f02a59c04d87bb5fe1b8008d8fd89260a coderank/dumps/repomap_tags-model-axum.jsonl
52
- b85e1bf8c68a9ff0a069fc35d8b901172f4ce9d6cdd90ea70e6a0d02072bbb9a coderank/dumps/repomap_tags-model-cobra.jsonl
53
- 288ac59738e652f50d3d3f2c640fd51f649454f1c4341af7ba1c8bab91c66308 coderank/dumps/repomap_tags-model-express.jsonl
54
- 679ac2ecfe940668709193003782e887b8278484d18a93b939203603af4caa37 coderank/dumps/repomap_tags-model-fd.jsonl
55
- d448ef3155927857cfd90ef32be430512e41c2dab61e1c33bedbcfabfacc28e1 coderank/dumps/repomap_tags-model-flask.jsonl
56
- e5b349569d88f7e2d29cfc33035b1b50c9df529fc10e2bca502b4d0cf6691b92 coderank/dumps/repomap_tags-model-gin.jsonl
57
- 53556264e7ee14cc7a84e5c4dfb736969e89e83c9440e69959ce201d99dcc3a6 coderank/dumps/repomap_tags-model-gson.jsonl
58
- e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 coderank/dumps/repomap_tags-model-httpx.jsonl
59
- 042607519a6f51d10b301c46307d562708461b2c06c1aeda63dc63004ac3f5ba coderank/dumps/repomap_tags-model-okhttp.jsonl
60
- a8c8f83ff7593156a95a317adc50375b080ab18a30321b33722b59a879d3545d coderank/dumps/repomap_tags-model-requests.jsonl
61
- d4e26d6f5eb13c5b3e65306c74f2f413e7f539af54fde4a5e564785c766b0de9 coderank/dumps/repomap_tags-model-ripgrep.jsonl
62
- 6642d6f9d816bd9ecb343de7b5d1265237286a79292e014fdb75da93e54397c2 coderank/dumps/repomap_tags-model-sinatra.jsonl
63
- 2cbe2dd858509e366c346b743c607a6cb34a591e26f6488c8a4952e798dda0ee coderank/dumps/repomap_tags-model-slim.jsonl
64
- aa766f7fab0130940bcdd273653b64caebbf71456c7823931a7ca7757c21c528 coderank/dumps/repomap_tags-model-thor.jsonl
65
- 1e1b341d493aa6cba1951b89570320ce59649192d3d1985f581adf2bd0274533 coderank/dumps/repomap_tags-model-zod.jsonl
66
- d4de787e189714deabdedf947cf35132a05bbd2d2c94f7e80e46a3e42718033a coderank/dumps/summarize_definitions-docs-axios.jsonl
67
- d134a1ba87d1e2853ad108fad077c953506fba00495cb7fe34bbdf3931bb6693 coderank/dumps/summarize_definitions-docs-axum.jsonl
68
- 817c853ccbe61f4e3da7c0aeab6ff103d18f8eaa6fbea67e9ebeefbc3857bf6e coderank/dumps/summarize_definitions-docs-cobra.jsonl
69
- 88a4302822ceb41ca6fc1e7bc2743e0ae25fe34b180df36d6377cb884ba0a416 coderank/dumps/summarize_definitions-docs-fd.jsonl
70
- 993c85a1c454add4f73beb087bf8ca983dd78fd648c7b58374a0ea472ad0df08 coderank/dumps/summarize_definitions-docs-gin.jsonl
71
- a36dac11920e5af69173534bfed697ffb3747f712ec34569faee02fb576236e7 coderank/dumps/summarize_definitions-docs-gson.jsonl
72
- fd9307f262b8c682e53397c48d89b588b910771ef0a57c378fe15b4ab203e43c coderank/dumps/summarize_definitions-docs-httpx.jsonl
73
- 755045282839ce07e85b1daaf542c599aaa722dec6a2bdfa125b124a9b755d56 coderank/dumps/summarize_definitions-docs-okhttp.jsonl
74
- 80c904159630f2925c9defcfb9fe81c0f596526a12c815d8fb282310c7436910 coderank/dumps/summarize_definitions-docs-ripgrep.jsonl
75
- 6352430b75ce4874c8d62cad2d406e84f342d4e2b483685a22fa9c536737fa99 coderank/dumps/summarize_definitions-docs-sinatra.jsonl
76
- 412176f1689a91e4ebc624ae72a09adf2f8c0e58a0fa4a7a14abbda19f0bc7e4 coderank/dumps/summarize_definitions-docs-slim.jsonl
77
- f87c6e2767f4f2a6abc3ce6698a9464c6e31590b0ac6d6a68243d18af49e0906 coderank/dumps/summarize_definitions-docs-thor.jsonl
78
- eb1f452a1689778938fd7bf6c638ea3ab9cd4229f41b5233169b55a67a209dcf coderank/dumps/summarize_definitions-docs-zod.jsonl
79
- eb73ad39e04f98f3b7e63e4e257bd3eb74188fca2cc72429c9d73c67f3906d59 coderank/dumps/summarize_definitions-model-axios.jsonl
80
- 504fb5db950d152ea824e0f1362d96bc77d444e21b666133b304d6f2b6b2ecb4 coderank/dumps/summarize_definitions-model-axum.jsonl
81
- 44dbd3241d5c23764f8382568a90a195d36b8d857b7433be154e07fabbe43121 coderank/dumps/summarize_definitions-model-cobra.jsonl
82
- be55a20fc30136536ac8b409bed5a2d6422b5d6f6c9b81125e18acbc3aced45f coderank/dumps/summarize_definitions-model-express.jsonl
83
- 4a432ad7c34be7a1b759935ce218c32ef2615231d0f851ba3d752f8c5f7e84ec coderank/dumps/summarize_definitions-model-fd.jsonl
84
- 8eb0901688c86c8f0c595e02c57eb6b63973ab5473bccf3804fe4191f584cdf3 coderank/dumps/summarize_definitions-model-flask.jsonl
85
- 3678538d2e621e5f04beb8cb764f9931583a03d141f5b4d98dcb2a64b5e95c97 coderank/dumps/summarize_definitions-model-gin.jsonl
86
- 2f41f8212572e84bfe3e6547fac6862e3ca8563ebae90c69b322d95c069939a6 coderank/dumps/summarize_definitions-model-gson.jsonl
87
- 5427da36c7f10eb4aeffa4b22b2dc56fbc8f022b753e519b94b359941f4da9cc coderank/dumps/summarize_definitions-model-httpx.jsonl
88
- 9e1c2de08296f90b4f61849b10330bfc2fa507e1721c82651b5c1ba978a3122a coderank/dumps/summarize_definitions-model-okhttp.jsonl
89
- f3d9b86b2cd59f2a93eee8f9aecf4c18472ec7d571f278da2d4f521964de0e0b coderank/dumps/summarize_definitions-model-requests.jsonl
90
- 39fef2476bc1d40a4a220ebafe383d5f332d4ea1399abf3eb9d405d3f5a46f28 coderank/dumps/summarize_definitions-model-ripgrep.jsonl
91
- 6d81a9d707e642eb1baacf99795c39278f8fb214b740a444a8f1fd7315d2880f coderank/dumps/summarize_definitions-model-sinatra.jsonl
92
- a40ee18ef6b1765728a7931150df5d58727204643518131f39868cb8f33dced6 coderank/dumps/summarize_definitions-model-slim.jsonl
93
- d53bbde3d99140eca323870fee03a9c8e23f3fe1c63c83fc3953fef283faeb66 coderank/dumps/summarize_definitions-model-thor.jsonl
94
- c77c2de2e51eaf60cccbdba6c80e6336d45e92bf28f916a39c1793089d092704 coderank/dumps/summarize_definitions-model-zod.jsonl
95
- 982a662c6c228be3d6411f9935015fbd2f209cb07b67a59a2af1e59c3180f0d8 coderank/dumps/summarize_structure-docs-axios.jsonl
96
- e463f81df1240c9541e9058e52e2d790b56745f4f2def695e4d4ebb9b847f8c8 coderank/dumps/summarize_structure-docs-axum.jsonl
97
- 05dc674b80797d26206444eb61e0e89837037ad66b7a44543537781b03c2e481 coderank/dumps/summarize_structure-docs-cobra.jsonl
98
- 0907291d079193c4bdcf1a15e189c287deeaa2cc5710eaabdbe1525e0d301eef coderank/dumps/summarize_structure-docs-fd.jsonl
99
- bf7d2e245155657138e1e7d1bcfed2efe861d9c57a114d8ad47702d69595a7a0 coderank/dumps/summarize_structure-docs-gin.jsonl
100
- 01036e8afcfa61bb268a532ccf6d66030076411a8f12eda15d429fcb608e899b coderank/dumps/summarize_structure-docs-gson.jsonl
101
- c4f2023db3d380c01f145fcde2e25a5f9b50371ab7178924360abc15cada2c45 coderank/dumps/summarize_structure-docs-httpx.jsonl
102
- ef4a91f7ce9d3c5b77fd843e11e85b79e6ae64e2728a0f5270b00e17f1ae2b05 coderank/dumps/summarize_structure-docs-okhttp.jsonl
103
- 569e705c1f9b9a97ece5bdd80b7a97625a22200eca76bef3ba2d0e891960f453 coderank/dumps/summarize_structure-docs-ripgrep.jsonl
104
- 43da310090a2d9991ceb05f9d478b714e2d0c64203f33a935d56d70eb3b3e6bc coderank/dumps/summarize_structure-docs-sinatra.jsonl
105
- 3d45525214e66953314f5afc8450c6fdd50b391cca4816582344bcc155633e94 coderank/dumps/summarize_structure-docs-slim.jsonl
106
- e02745dc7af454543d83c8b1955077d57e3ebb0247e52fee70aacdab2c4b0281 coderank/dumps/summarize_structure-docs-thor.jsonl
107
- 9607afc3ee04a1fdd2600fea082730ee990c7e3ebc24e8f8d7aab13c7ae0841e coderank/dumps/summarize_structure-docs-zod.jsonl
108
- 29d39e366c4cdf9e0b328eef1af4de50c55571471fbf0eeb3d72f895013f1466 coderank/dumps/summarize_structure-model-axios.jsonl
109
- 022a604b69e6e8c63cf9d1ab5d55f5073b5c8c6ed6d05ef9a2411f303c5d2794 coderank/dumps/summarize_structure-model-axum.jsonl
110
- f8d572042f14a9a9946e2933f2a6beab51c993c8f4b26179fa83f75e63e53190 coderank/dumps/summarize_structure-model-cobra.jsonl
111
- b534e0995e7f0781665787b2086fd50c2a4a45873cf63d7c297f1e828c357328 coderank/dumps/summarize_structure-model-express.jsonl
112
- 60d21005597530635b01c020dd308acf6c5464e74affc67e33f69654d0af6770 coderank/dumps/summarize_structure-model-fd.jsonl
113
- cf121de2c1493634573e67e86668e0541448a3366da6b1afc15689d80c7d7034 coderank/dumps/summarize_structure-model-flask.jsonl
114
- d3ebc36a3be4ce4685245ba8f11c5e46401a6c2ae9cf905117218bcc412d3546 coderank/dumps/summarize_structure-model-gin.jsonl
115
- d3795241fa1c3445466775cfcf6ac86b559825454cc2e64ac04da9e9efec82ca coderank/dumps/summarize_structure-model-gson.jsonl
116
- f824225456e48888fb39aa238108e6a06cf1786a5cb405ad7c5108fe84cd41fd coderank/dumps/summarize_structure-model-httpx.jsonl
117
- df04fb09c053ee6b20501dacfe2829a86f0eb3fae9b58b6efd3939844db2e382 coderank/dumps/summarize_structure-model-okhttp.jsonl
118
- f2aca8248d0dfa6b2951e3d6027043898e0262a1a5f79221b7d48c7aaca17222 coderank/dumps/summarize_structure-model-requests.jsonl
119
- d2e9cedc4a2ee47f78bdde1d61bc30faa1766adc8d51a9dd7d9b83dcb2725830 coderank/dumps/summarize_structure-model-ripgrep.jsonl
120
- 2733680220b158e7500c411cebdeeb03e38c98747528b22079e629dc6d2cd9e4 coderank/dumps/summarize_structure-model-sinatra.jsonl
121
- ce1fe65ce345e6dee0de309b140c8dcd9aa857ba028e307864a21f0762375398 coderank/dumps/summarize_structure-model-slim.jsonl
122
- b272291b1f0e4b0aa4fca80d5675918dea18814a33d3568de9e83376c5e701df coderank/dumps/summarize_structure-model-thor.jsonl
123
- 6b5e7678c24d7fd7175802a5608a468916f17e02773a33a02660c50ee58867b0 coderank/dumps/summarize_structure-model-zod.jsonl
124
- 998c6d5eedc8c7443f3de5650007a8cb06d43cc4399d246af43d9561c47301e4 coderank/pairs/docs-axios.jsonl
125
- 530ab0dbb7109c750d2bd46e09caa143e13261442f343f7bc55451b26942f8f3 coderank/pairs/docs-axum.jsonl
126
- be739e8b08ec8a9eb37f84ac1e50e7052b90855f60df57f89a45d2c27b05b62c coderank/pairs/docs-cobra.jsonl
127
- e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 coderank/pairs/docs-express.jsonl
128
- c9277637cf12e41c3bcbdb248d77b13ce8b3e4703dd124ed97b0d259a1566466 coderank/pairs/docs-fd.jsonl
129
- e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 coderank/pairs/docs-flask.jsonl
130
- 77f196de917b9056f37c51aa506dd99f73e627d306633864bcabf48e19e082bf coderank/pairs/docs-gin.jsonl
131
- 089466b11db11704ae4bbcf51b16eaee28b152f3374350fa2abc1df51b3c1548 coderank/pairs/docs-gson.jsonl
132
- 4585210652c4ae8920fb3627deb6762de7c9b3fc91e482b7a3e738d98e3f07d1 coderank/pairs/docs-httpx.jsonl
133
- 6e3b828499f84eeef1a16d3563f29c6ef3e45b9d1248ee03a23e4a54ed5ddf4a coderank/pairs/docs-okhttp.jsonl
134
- e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 coderank/pairs/docs-requests.jsonl
135
- f4aed9ff9ba9b581b4f8c852821d6e3f14e2291da3c895da01e2bf33aa7a886b coderank/pairs/docs-ripgrep.jsonl
136
- e04687b8333e1d8aceca7f9599d0092a56c6f2d32ba4a0f72e6b8f1925bb4047 coderank/pairs/docs-sinatra.jsonl
137
- b9178f3efa0df34e16cf2c32e9424ed55e9aa9a4306368b3719ea40f4b1ec551 coderank/pairs/docs-slim.jsonl
138
- 4f40152e46b8ca5e5d6540e567a36ef1f94493130f61a05ae7e8a26dc5c47427 coderank/pairs/docs-thor.jsonl
139
- cc412c5833ee9e28f4b92058c4605fd108c0fda7196317a3c5c0843c4f8ec302 coderank/pairs/docs-zod.jsonl
140
- 8f12d3deb77a884f577ffa10bbb4628260c71c5e2be8c1ace6cf8a5f4bd10503 coderank/pairs/model-axios.jsonl
141
- 7fc85185b21b26ade374034ee55a3b62b5c9abf7b0ec8f33106957292d5ba18f coderank/pairs/model-axum.jsonl
142
- 767cbf2ca31e728aa6dde8080cb8eddff6273688d213adc46d5bbad40d695664 coderank/pairs/model-cobra.jsonl
143
- f7d7e19f2d6f06b8eca2c0223f3c43dce3eca1370ba6a2517a6b0801501d7e8c coderank/pairs/model-express.jsonl
144
- 0f486dfa2d340155f96b9c3e1ef8060dadee0a9c6d415c96c5bf13cf8abf7ca9 coderank/pairs/model-fd.jsonl
145
- 25bedb286f7b2dbb5025b93f6e11f3de3f9f487027d160fc1f14211c89c5dc91 coderank/pairs/model-flask.jsonl
146
- 1451daabf55fc3a83544fcc3e28d64dc91f361d15e3955852d2e148f26184e11 coderank/pairs/model-gin.jsonl
147
- e674e208bc303e7fa89de0b5454f837e13d82b07623b6e0db34754669f6382ad coderank/pairs/model-gson.jsonl
148
- 99f4aef319a2926698674859265f8a13f8c4adc7b754077421f53a5441891a47 coderank/pairs/model-httpx.jsonl
149
- 9b0a61541a15586f85e1f810a5d1ee7c55e78cc50e14050370be9bb0ef7db809 coderank/pairs/model-okhttp.jsonl
150
- d452b47569a1721d17323a2a7b33fde102f63abaac9176ae47a1d48963dd907a coderank/pairs/model-requests.jsonl
151
- aecb05e06219fa1274cc76214864bcb041acefa0c6bf3b3c60c510bdb9eeefd6 coderank/pairs/model-ripgrep.jsonl
152
- 804e473f4d7d940241ed55d18b736df91501a15b0df9de566a89848d04ce5a6d coderank/pairs/model-sinatra.jsonl
153
- 06e31042ffca6f1105f182d0c652efe98b9dc4fbb33b62cdef7d0d3acbb71f69 coderank/pairs/model-slim.jsonl
154
- 61612faa4d87a194d40ccf95a96665d10cf986ae5eb0153f4851c06b4ee94c9c coderank/pairs/model-thor.jsonl
155
- 66abd50a893a931bf9c00a774d07e81e122fad83b025da2722880c20a770e644 coderank/pairs/model-zod.jsonl
156
- b33046f69b6d1951f5389c0720f5b0ce8687228515dbc688a4b8f6b7e7bdd7a0 coderank/sites/rows-sites-axios.jsonl
157
- 540be4c128810f3a0e85cf0b1b71b05a55a32715692715d83a54892ac20c6c4d coderank/sites/rows-sites-axum.jsonl
158
- 9b0da6f328d72497abbed61161de1a429664579c388f31c8e6ff91ed939f06ec coderank/sites/rows-sites-cobra.jsonl
159
- cf6563e0af829f92bb106f1a5fd196a97711902d90b6570e530c55c9b9f8b889 coderank/sites/rows-sites-express.jsonl
160
- 609d47a13956a1c928fb3185190146ab3ea0a89afb5aba26c20044e7d77775d8 coderank/sites/rows-sites-fd.jsonl
161
- cf4de7b99baa7dcbacf26e337d583cba29433678133d2f6540e85707e788f22e coderank/sites/rows-sites-flask.jsonl
162
- 12d4d08025732b9ee5e0a2bdd7d0ce906c98d0732e03ff3f1cbee0168a939224 coderank/sites/rows-sites-gin.jsonl
163
- 9f45410c3e68e187d636eabd2aa195286b5bf01e33b01f63fbacd683c8c92389 coderank/sites/rows-sites-gson.jsonl
164
- 45c13cc83c54419f1e57bbedac4377dddb84b1c779e1c838634bc01787365be2 coderank/sites/rows-sites-httpx.jsonl
165
- 37ad7a288e1d0e8fe92dfba0ed0181e7b0afd56b5104aebd208d99d127028e6b coderank/sites/rows-sites-okhttp.jsonl
166
- 1c2964d79f32b8494bcb76d4f07a93dbb130dcfe2278440d536c9b78f7c599ab coderank/sites/rows-sites-requests.jsonl
167
- 7d5fa209fcb688d9b05b02aa13503869d5f63a3517fbe463f21f88ae05c85cd8 coderank/sites/rows-sites-slim.jsonl
168
- aae839ef2eeb2188c85015f338cc0ed74798dc925b5118c54c1fd45b3e843fce coderank/sites/rows-sites-thor.jsonl
169
- 9f889775e9a19377d060bacc09f7a11bb09ed50acbf523d425e392fbcdcd5f99 coderank/units/axios.jsonl
170
- f84ef9bd37d373f3117aaa7c256b581ab8670697030a2c145cd9b4680a96a380 coderank/units/axum.jsonl
171
- 2932a270c2b5b76d7a4dd699fa569945db4bc0b4bc549110545dcb684f95485e coderank/units/cobra.jsonl
172
- 7ce8be3a2bea8553a163924cb5577279b4ff0d2a00389fd747568f4dd27e2edf coderank/units/express.jsonl
173
- e2e4cb29127163aea036ee1139d88225f09858597507553063b05f1e2a91cdb7 coderank/units/fd.jsonl
174
- 8eccd2067eea7427e6550659653a93a9d883ef531159b8ac806cf2c8fbbdff1c coderank/units/flask.jsonl
175
- a5e44ede5d75cbd1f9ae6bf829001717fac12c59741efdb4acffada3d1eb5f8f coderank/units/gin.jsonl
176
- 6338623a4512d197eb8e2d94da21d139923ff3cb5663ff665c24605fabf26265 coderank/units/gson.jsonl
177
- d8de37959c02c47055e6fc4cb8194ed048fe611811672e3e4bd02a63150531b9 coderank/units/httpx.jsonl
178
- 2eb89801e5093e44f49e615e2a3738d20a8f0a89d437227ab2f84c2b1dcc0466 coderank/units/okhttp.jsonl
179
- ea2815d1250b86879a36dc4f62b9f30a87f8ca45159ac3287ef7fbdf70dda28b coderank/units/requests.jsonl
180
- b22d4bf5fd1a7eb8916b4f44bc71bb3295372b06905b9fbc62c2ff5072cce30d coderank/units/ripgrep.jsonl
181
- 96a59e2627dc8e546d87dc795f9fd1598297051fc4065991a1cd43becfc52b2d coderank/units/sinatra.jsonl
182
- 8833fcbc679766581cf2e0d2f4afa6067593c420a607f5b5a822181ac54df309 coderank/units/slim.jsonl
183
- c2b13a10ad583904e26a43188e9fad06c508fafe15bf17b8399227a7aadf7a71 coderank/units/thor.jsonl
184
- 977460c1f255e340e858347ece342efdb0ae458f10a100fdf93c871840dc0045 coderank/units/zod.jsonl
185
- c77abd8b40965c05cc43a7545e8c2db2405bdcfc18a73bad254632d523b67bcd corpora/independent.json
186
- 51e3a65258fc34581bc8cfaae0f7c97f444f0a50f3dda906c77b6554fd5c66d3 corpora/original-open1.json
187
- 51e3a65258fc34581bc8cfaae0f7c97f444f0a50f3dda906c77b6554fd5c66d3 corpus.json
188
- 6381bd724d7c122265d79a092d0529e15ebdd0708dc6510e8f942c115278c6a0 evidence/code-rank-provenance.json
189
- 80bb261c8433594f12508a60f7a859df01b526e8a7ff47993e817256a72b4b4b evidence/preparation.json
190
- 83483e997982d517385a1f884f835d601dea278d09c85f154b995aedc0aa4f41 evidence/sessions-provenance.json
191
- 83124df425cda00df69ea82c9b4308310d6c378b915edb8a448adaeebdb6d3b0 evidence/split-identities.json
192
- 3cac53164d5a4c4f81e3fdce6001b08c8cf13e2a739ad08558cd2baa9b227bef historical/holdout.jsonl
193
- 4923f345359a0df64513d6cd2f689483df5b1a3081734760c2fbaf9235df7fe0 historical/open1-judged.jsonl
194
- 0fea3a5ab5eee75148c57252b062f1bbfa934bfbe561b3c1a00ac03f8fec137d historical/train.jsonl
195
- d0c39770ec9776849c42adc9e4c70b64c2463574b8e306831694279e68ca1bd6 historical/val.jsonl
196
- f50ec8ba164e5b01536b02c5d62c81285dee3732822b1381e39a69b869ed80c9 holdout.jsonl
197
- 82761059eaedacb3356803aea8a170d8298609f91b14fc32ee1bfb40d690183c licenses/axios/LICENSE
198
- 008c87afcd2e626eaf564093250bed06dd7efb5732113264bba3dda8f1c556a1 licenses/axum/LICENSE
199
- 5e3400b93bbb099e83e52bab885e7441750673c21f97988ca3f1240639b63283 licenses/cobra/LICENSE.txt
200
- 95a5762890e5c1c9808921cef095661fc482c5e1f0bba31446ac85595df6237c licenses/express/LICENSE
201
- 73c83c60d817e7df1943cb3f0af81e4939a8352c9a96c2fd00451b1116fa635c licenses/fd/LICENSE-APACHE
202
- 322cfc7aa0c774d0eca3b2610f1d414de3ddbd7d8dd4b9dea941a13a6eb07455 licenses/fd/LICENSE-MIT
203
- 489a8e1108509ed98a37bb983e11e0f7e1d31f0bd8f99a79c8448e7ff37d07ea licenses/flask/LICENSE.txt
204
- b104efb2c7700691650f27034e8541c5ae0ed9af54d884287f19a7467ca2fe7f licenses/gin/LICENSE
205
- cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30 licenses/gson/LICENSE
206
- 4ec59d544f12b5f539a3a716fd321ac58ccd8030b465221f2c880200cdf28d8d licenses/httpx/LICENSE.md
207
- cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30 licenses/okhttp/LICENSE.txt
208
- 09e8a9bcec8067104652c168685ab0931e7868f9c8284b66f5ae6edae5f1130b licenses/requests/LICENSE
209
- f5110972dedad2b4e9d314518daf3b7d72d6e02e499acd802181de6f74571dcc licenses/requests/NOTICE
210
- 01c266bced4a434da0051174d6bee16a4c82cf634e2679b6155d40d75012390f licenses/ripgrep/COPYING
211
- 0f96a83840e146e43c0ec96a22ec1f392e0680e6c1226e6f3ba87e0740af850f licenses/ripgrep/LICENSE-MIT
212
- 7e12e5df4bae12cb21581ba157ced20e1986a0508dd10d0e8a4ab9a4cf94e85c licenses/ripgrep/UNLICENSE
213
- f2f60556a2d5c8d7ef9d54ec1c62152ff40d626b807aba5bdc5126f669ed500f licenses/sinatra/LICENSE
214
- 24d84675cc9fffd507101fd43fecb5dd7f87679f065d90b2ef597d0a8b7e5994 licenses/slim/LICENSE.md
215
- 4df3e92a124b84d6767b1a0f1302600ea41d7616cda0b0fc40c1eefe4c9e414b licenses/thor/LICENSE.md
216
- 3f1189b28e3866e0d979968d466b78f813f76827cfdca1fbb124cc0a5c8841f8 licenses/zod/LICENSE
217
  260f26ccdd53437f1d01b269d9f510dde1b497b9b64a45a7807d83231a2a7bb5 row.schema.json
218
- 65752a429f324f36e28ba17e721d3fe2853b010b6a44cdb921d8a18592a2f62c sessions/combined-snapshot-rows.jsonl
219
- b886be4e60b94a9ae27a6e36fe60ad2aafcf2ac8d574170dffd79fded77ac152 sessions/final-collected-rows.jsonl
220
- a24c3a26bcee3f5e9c1af79c447ea38866e2ef2d4d2a03da9452f0e0bb92f240 sessions/later-rows.jsonl
221
- 6d8af5d81cbd81874f5a4d6ce1df9dad1eba9cd66e8ee534f01057a1cb41eae5 sessions/live-snapshot-rows.jsonl
222
- 59b16f6b4d9668600ac499e30fc4f44e3ac58ca57dbf1d05f48388cf3d2d8369 sessions/transcripts.jsonl.gz
223
  23a4abda8e51e7fc821615b5a83dc250d115d702cad19da4cf68863e260b1d43 tasks.jsonl
224
- ec3a0008eb903cd455912ce094a1e53884498a77ef76a20fa72805ab7b9aaeb5 train.jsonl
225
- 5cd33f4ee6326f307acc0e0d0225a6367859dd85227132633de394952b2a57da training/code-rank/recipe.json
226
- 2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54 training/code-rank/scripts/artifact_paths.py
227
- 4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f training/code-rank/scripts/decide/bench_latency.py
228
- c4594e1514fffb0b64b931476c37890ce872259babb28aeaadc934e737a8c607 training/code-rank/scripts/decide/calibrate.py
229
- d0c7dda2d181bb81646e61090a971ebc1c5922703f00170572ed2ce40ce8652b training/code-rank/scripts/decide/corpus.py
230
- f4ca003205839d391c10d26417af8239b86b8af8471f30969441346edcf770b5 training/code-rank/scripts/decide/headfile.py
231
- b574de47f0deed015f59d71cbfd50130eaf6a40c9a54c3de261a66d6b0f4df70 training/code-rank/scripts/decide/parity_probe.py
232
- f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98 training/code-rank/scripts/decide/rank-rubric.json
233
- 4a2dc07a0984294030c7691c9727ce98392e8ef578fbfa491ae7945635577172 training/code-rank/scripts/decide/replay_eval.py
234
- 60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d training/code-rank/scripts/decide/rerank/synthesize_sites.py
235
- b071fa53ed417afc7c8a1779460d60950d30f8472daf4ef4eb20a976815e6f9a training/code-rank/scripts/decide/rerank/train_rerank.py
236
- 260f26ccdd53437f1d01b269d9f510dde1b497b9b64a45a7807d83231a2a7bb5 training/code-rank/scripts/decide/row.schema.json
237
- 9d9ec7c18661ffa53820f272b2324fec4a1d72ee578ca5f0df22ae5bda364b17 training/code-rank/scripts/decide/rows.py
238
- e9343227428319b8c3e75bf8de2976f78fefd9f3720ce3cfdd4a3c840c456772 training/code-rank/scripts/decide/train.py
239
- cd648d84da48acdc5312a3affd846a93bda39d7a6c94dbd1f446ec77b8872dd9 training/turn-load/requirements-resolved.txt
240
  2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54 training/turn-load/scripts/artifact_paths.py
 
241
  4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f training/turn-load/scripts/decide/bench_latency.py
242
  6cef23a6e54fdb048e223cb336ca0d4f32168b28190208259befdb8cfcceac2b training/turn-load/scripts/decide/calibrate.py
243
- a8b8295261f19dda992c81fdec8aab8fbc2c065bc28cd186f03def6c771c7e8c training/turn-load/scripts/decide/corpus.py
244
  e6746d919b5aa9217d1e666b97dd09c807442c31d24fb04654dc22035a4014cd training/turn-load/scripts/decide/headfile.py
245
- ef4263b611d769752b0284d0ed17c538e0ae04a2652969b22f851584132a3d6f training/turn-load/scripts/decide/parity_probe.py
246
  f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98 training/turn-load/scripts/decide/rank-rubric.json
247
  9fe36ccc8021cc62666630ac0f41ecf5c1d3a0b8a24ad438461f8120a9eb6228 training/turn-load/scripts/decide/replay_eval.py
 
248
  60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d training/turn-load/scripts/decide/rerank/synthesize_sites.py
249
- 1fed73f53fd2e3512ec0d626edd3b41bb0969bf52ca302dd1fc4c772737f6e59 training/turn-load/scripts/decide/rerank/train_rerank.py
250
  70d83cc6d22bd465cad54c696403be9bbfd7152460d74ee7ed9462046f377913 training/turn-load/scripts/decide/row.schema.json
251
  2554a171844b7a432920b82a0a9a7d070fac6e7777f171e9c53b8eed557a325e training/turn-load/scripts/decide/rows.py
252
  cb9bd0edd5472ba114310ba9f50a3d2089b2964696250ebea1c2ca9f0f0037e7 training/turn-load/scripts/decide/skill_discovery.py
253
- 18061510ed0299786f812f7b41d542ab4469d0d6148e9d93eb23bd7582d4b3b5 training/turn-load/scripts/decide/train.py
254
- 0298ad3382a82c6138d48ce4d81aaa9384ebb755f775d802ff0ae7a8766b6685 training/unit-rank/mix.py
255
- 9fea6d672870b24a829325d0c7d13ccceaf8927de35fe2af169b087d5653179b training/unit-rank/prepare_rank_data.py
256
- 4689a183766b25872744f0bc712e923560324572a25ab07a4a337104e7d80654 training/unit-rank/requirements-cuda.txt
257
- 2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54 training/unit-rank/scripts/artifact_paths.py
258
- 4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f training/unit-rank/scripts/decide/bench_latency.py
259
- 6cef23a6e54fdb048e223cb336ca0d4f32168b28190208259befdb8cfcceac2b training/unit-rank/scripts/decide/calibrate.py
260
- d0c7dda2d181bb81646e61090a971ebc1c5922703f00170572ed2ce40ce8652b training/unit-rank/scripts/decide/corpus.py
261
- e6746d919b5aa9217d1e666b97dd09c807442c31d24fb04654dc22035a4014cd training/unit-rank/scripts/decide/headfile.py
262
- b2ea43b655b6dceb637ab7fbee9ff3755c6cf34b976cfeadd6119a0734e2a9b9 training/unit-rank/scripts/decide/parity_probe.py
263
- f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98 training/unit-rank/scripts/decide/rank-rubric.json
264
- 9fe36ccc8021cc62666630ac0f41ecf5c1d3a0b8a24ad438461f8120a9eb6228 training/unit-rank/scripts/decide/replay_eval.py
265
- 60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d training/unit-rank/scripts/decide/rerank/synthesize_sites.py
266
- 1fed73f53fd2e3512ec0d626edd3b41bb0969bf52ca302dd1fc4c772737f6e59 training/unit-rank/scripts/decide/rerank/train_rerank.py
267
- 70d83cc6d22bd465cad54c696403be9bbfd7152460d74ee7ed9462046f377913 training/unit-rank/scripts/decide/row.schema.json
268
- 2554a171844b7a432920b82a0a9a7d070fac6e7777f171e9c53b8eed557a325e training/unit-rank/scripts/decide/rows.py
269
- cb9bd0edd5472ba114310ba9f50a3d2089b2964696250ebea1c2ca9f0f0037e7 training/unit-rank/scripts/decide/skill_discovery.py
270
- a70bd2179da44b6d9356585f3b86674556c85aec5cfb9b87e2958af0d180fd4f training/unit-rank/scripts/decide/train.py
271
- de4c3becb1dbeeaa208cba89cf229a8150781deabe0034f227256c0d5eb57503 unit-rank/generated-train.jsonl
272
- 5fda6375ca6d6ffeb08fd8d20e7c2342c5051101ab77d52231178e892093f745 unit-rank/historical-acceptance.jsonl
273
- 20a3ebd2501f6b4ad092827eca2d9de79e477847d0ff0718f13adc87d7854ed9 unit-rank/selection.jsonl
274
- 5c857914f92211be68a6c9b8087c5589155c48ef0182c986484c5588f43b999f unit-rank/train.jsonl
275
- a2f446c6b3b34b2dec1b70a04836ebc55e7951bcc4bcca0556019cb030dcda30 val.jsonl
 
1
+ 7e4b81e09541d632ef464c71d291659ab7bdb869b56c45ba8c3e52f84e44d003 FILES.json
2
  3c2307bbbbb88f79560a165b098c600da8f20fb1ec024586d4be192ebc9c6f8e LICENSE
3
+ 1daf6121748acc99951952c52db5e3f39fa97be914a982d671e03e2d060b830e NOTICE
4
+ 1aca8e2b351629342f6297f3d5dcc77bee8f3060b84085950e3b175b93cfff66 PROVENANCE.json
5
+ 6d83cb155ee60b174fbbad6411eb67a3d5c960fc23ca307e515a07a0df47d9b1 README.md
6
+ 429ef9be68f483c275773cb136510a6e41a71268c10e3febad302aa8ad212ecc TRAINING.md
7
  7e09a014dd94f32d95196a3531e046c1904d526a593b670f448f1d87168a44ff agreement.jsonl
8
+ e00fb65172e65a466248ac0a8ff6985740e8762e04fb412a74c44eadcf30b6ac corpora/independent.json
9
+ af380a5338cbdd5652bbbee97e99f7317ebf4e32f766f4ad6c7862a51b4a74ea corpus.json
10
+ 6ac4d873ce4802e10fc6178c4511508e36b88e86744335f0c5c6d9c23751fb4d holdout.jsonl
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  260f26ccdd53437f1d01b269d9f510dde1b497b9b64a45a7807d83231a2a7bb5 row.schema.json
 
 
 
 
 
12
  23a4abda8e51e7fc821615b5a83dc250d115d702cad19da4cf68863e260b1d43 tasks.jsonl
13
+ 12c4b2d46ee1088882be996f409b661c2b9741c5e97cd6a71e3d580102633685 train.jsonl
14
+ 7a6e4ad8d2294a03cb2164090d41b0bfbd75e6ed1b4b99fb3143221c6b5ed23c training/turn-load/guide_audit.py
15
+ 73a1a984be21a67a0c9ba320ed4d8563137596a9c88d0653145edde511926dea training/turn-load/relabel_guides.py
 
 
 
 
 
 
 
 
 
 
 
 
 
16
  2b27e5d68e85cf94176fd90b46c51b224e193812374ef5c53d489d554093fd54 training/turn-load/scripts/artifact_paths.py
17
+ 35803a2ec2b104cd90f5628129675bb1145c4dfc1173a4ab2874a00542315021 training/turn-load/scripts/decide/README.md
18
  4d3ba4e7b59adb634a24928491eb36bfd9e9df4de18a43a2d80dd9a042a8117f training/turn-load/scripts/decide/bench_latency.py
19
  6cef23a6e54fdb048e223cb336ca0d4f32168b28190208259befdb8cfcceac2b training/turn-load/scripts/decide/calibrate.py
20
+ 5fe3241527d042ce948c24f18b9e87a3bee8fa1e86753102074efb7c88b916f5 training/turn-load/scripts/decide/corpus.py
21
  e6746d919b5aa9217d1e666b97dd09c807442c31d24fb04654dc22035a4014cd training/turn-load/scripts/decide/headfile.py
22
+ 6b28eea348fde43bd540a4dbbe481ee91718a1895597012023aebb5147b19d9c training/turn-load/scripts/decide/parity_probe.py
23
  f4accce5b0cabcaf5f4e8d51b7574b134963cff05f5fc8bc0ac635092596fd98 training/turn-load/scripts/decide/rank-rubric.json
24
  9fe36ccc8021cc62666630ac0f41ecf5c1d3a0b8a24ad438461f8120a9eb6228 training/turn-load/scripts/decide/replay_eval.py
25
+ 99cd5f37e38c040ca5145a92578cdcd1e54d5a1f5c237d95aa0316ba823d9340 training/turn-load/scripts/decide/rerank/README.md
26
  60ac41772c40089453ebb3450e90e1f97a0dd2efd7e81fc5af287760bf23913d training/turn-load/scripts/decide/rerank/synthesize_sites.py
27
+ 92557e17b99a94783e00eeca43bcb5365e232578e8b2878768e0fea1232ff065 training/turn-load/scripts/decide/rerank/train_rerank.py
28
  70d83cc6d22bd465cad54c696403be9bbfd7152460d74ee7ed9462046f377913 training/turn-load/scripts/decide/row.schema.json
29
  2554a171844b7a432920b82a0a9a7d070fac6e7777f171e9c53b8eed557a325e training/turn-load/scripts/decide/rows.py
30
  cb9bd0edd5472ba114310ba9f50a3d2089b2964696250ebea1c2ca9f0f0037e7 training/turn-load/scripts/decide/skill_discovery.py
31
+ a2ab2243213d918439d577eba9d0fd87c135aa6eee2bc0a647de077ac04a143d training/turn-load/scripts/decide/train.py
32
+ 7f500d26f5d9db5c6a02b6e3e4d03df27b40dbdeaf9b1853b8196a7fdab7afc2 training/turn-load/train.sh
33
+ da49f03dc149ee0ea7ccddf582001973a8ade51dce0e77d857738ed960cb1c93 training/turn-load/turn_probe.py
34
+ 06cc4a0225da9d167662942175efb63aa065d2da3ac5058ad9b6045096b8c051 val.jsonl
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
TRAINING.md CHANGED
@@ -1,73 +1,15 @@
1
  # Training from the archived inputs
2
 
3
- Use the separate trainer snapshots under `training/turn-load`,
4
- `training/unit-rank`, and `training/code-rank`. The `scripts/decide` layout is
5
- preserved for their sibling imports. Install the compatible PyTorch, Laya,
6
- Transformers and safetensors dependencies in an isolated environment. The B5
7
- resolved environment is recorded in `training/turn-load/requirements-resolved.txt`;
8
- E4's requirements are in `training/unit-rank/requirements-cuda.txt`.
9
-
10
- The backbone is `convaiinnovations/laya-multilingual` at revision
11
- `e4e9ddf21a7b1903b7acffd8814ad4307bf63a67`. Obtain that revision explicitly rather
12
- than resolving the moving default revision. The historical loaders consume the
13
- model identifier/cache; configure the environment to resolve the pinned snapshot.
14
-
15
- Run commands from this dataset directory. They describe the archived recipes;
16
- release preparation did not execute them. Choose output paths outside the sealed
17
- release directory.
18
-
19
- ## B5 turn-load
20
-
21
- Use `training/turn-load/scripts/decide/train.py` with:
22
-
23
- - `--corpus corpora/independent.json --train train.jsonl --val val.jsonl`
24
- - `--model convaiinnovations/laya-multilingual --families tools`
25
- - `--tool-truth consensus --tool-weight none --tool-negatives 24`
26
- - `--pos-weight 6 --lr 0.0005 --seed 11 --batch-size 64`
27
- - `--epochs 45 --patience 8 --label open1-turn-load-B5-release-independent`
28
- - `--out <output>/turn-load.safetensors`
29
-
30
- The independent encoding and 60-word tool options come from the corpus. The
31
- recorded backbone context is 1,024 tokens and head option budget 512 tokens;
32
- these are not interchangeable. Negative sampling uses inverse sampling weights.
33
-
34
- ## E4 unit-rank
35
-
36
- Use `training/unit-rank/scripts/decide/train.py` with:
37
-
38
- - `--corpus corpus.json --train unit-rank/train.jsonl --val unit-rank/selection.jsonl`
39
- - `--model convaiinnovations/laya-multilingual --families skills,requests`
40
- - `--rank-levels skills-blended --skill-scored 8 --skill-zeros 12`
41
- - `--seed 11 --batch-size 32 --label open1-unit-rank-e4-dense1`
42
- - `--out <output>/unit-rank.safetensors`
43
-
44
- Other options retain the included trainer's defaults. The archived mixed input
45
- is supplied directly. To inspect its derivation, `training/unit-rank/mix.py`
46
- accepts output, fraction `0.5`, seed `7`, `train.jsonl`, then
47
- `unit-rank/generated-train.jsonl`. Do not overwrite the sealed input.
48
-
49
- ## Open1 code-rank
50
-
51
- Use `training/code-rank/scripts/decide/rerank/train_rerank.py`. Supply one
52
- `--dump` argument for every file, in order, in
53
- `training/code-rank/recipe.json`'s `input_files` array. Keep the order: the seeded
54
- shuffle and candidate sampling depend on it. Then supply:
55
-
56
- - `--model convaiinnovations/laya-multilingual --label open1-code-rank`
57
- - `--epochs 6 --batch-size 64 --max-len 512 --seed 7`
58
- - `--out <output>/code-rank.safetensors`
59
-
60
- The recorded recipe excludes empty inputs and filenames containing zod, sinatra,
61
- or ripgrep. The trainer splits the selected row pool 90/10 before constructing
62
- labeled candidates. The reconstructed training-example count matches the saved
63
- head's 251,360 examples. This count is an input-lineage check, not a training run.
64
-
65
- ## Sessions and original judging
66
-
67
- `tasks.jsonl` preserves generated prompts and outcomes. Session exports retain
68
- message IDs so requests can be joined to `opening_message_id` and session IDs
69
- in the decision rows. Every session referenced by the training, validation and holdout rows has a
70
- retained conversation export; coverage counts are in `PROVENANCE.json`.
71
- The original split and judge agreement files remain under their historical
72
- identity. The present factory can generate new sessions, but doing so is a new
73
- model run and is not a deterministic replay of every archived event.
 
1
  # Training from the archived inputs
2
 
3
+ open1-b7g-e4 derives from open1-b5-e4: the same rows in the same order, with `labels.guides`
4
+ relabelled from tool calls (`training/turn-load/relabel_guides.py`). Train the B7 and
5
+ B7G heads with `training/turn-load/train.sh`:
6
+
7
+ - `train.sh GPU B7 RUN`: `--families tools,guides` on `corpora/independent.json`,
8
+ `--tool-truth consensus --tool-weight none --tool-negatives 24 --pos-weight 6
9
+ --lr 0.0005 --seed 11 --batch-size 64 --epochs 45 --patience 8`.
10
+ - `train.sh GPU B7G RUN`: the same with `--families guides`.
11
+
12
+ Score a head on every tool and guide option with `training/turn-load/turn_probe.py` and
13
+ audit omissions with `training/turn-load/guide_audit.py`. The backbone is
14
+ `convaiinnovations/laya-multilingual` at revision
15
+ `e4e9ddf21a7b1903b7acffd8814ad4307bf63a67`; obtain that revision explicitly.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
corpora/independent.json CHANGED
@@ -6,17 +6,21 @@
6
  "head_tokens": 512
7
  },
8
  "questions": {
9
- "deadline_ms": 2500,
10
  "tools": {
 
11
  "question": "Which of these tools will completing this request call?",
12
  "option_words": 60,
13
- "load_at": 0.43,
14
- "independent": true
 
 
15
  },
16
  "guides": {
 
17
  "question": "Which of this guidance does completing this request need?",
18
  "option_words": 8,
19
- "omit_below": 0.2,
20
  "confidence_floor": 0.5
21
  },
22
  "kind": {
@@ -30,23 +34,17 @@
30
  "inspect": "reading and understanding the project without changing it",
31
  "run": "building, testing, running, or verifying something"
32
  }
33
- },
34
- "skills": {
35
- "preload_at": 3.4,
36
- "list_at": 1,
37
- "roster_max": 6
38
  }
39
  },
40
  "request": {
41
- "deadline_ms": 2500,
42
- "load_at": 2.3,
43
- "max_loads": 12,
44
- "nearest_loads": 3
45
  },
46
  "lookup": {
47
- "deadline_ms": 2500,
48
- "list_at": 1,
49
- "max": 8
50
  },
51
  "kinds": [
52
  "answer_only",
@@ -116,6 +114,7 @@
116
  "floor": [
117
  "answer_decision",
118
  "ask_user",
 
119
  "edit",
120
  "extend_worker_budget",
121
  "grep",
@@ -161,6 +160,7 @@
161
  "floor": [
162
  "answer_decision",
163
  "ask_user",
 
164
  "extend_worker_budget",
165
  "promote_overlay",
166
  "recall",
@@ -205,6 +205,7 @@
205
  "command_output",
206
  "command_stop",
207
  "copy",
 
208
  "delete",
209
  "diff",
210
  "edit",
@@ -288,6 +289,7 @@
288
  "command",
289
  "command_output",
290
  "command_stop",
 
291
  "extend_worker_budget",
292
  "fetch_url",
293
  "grep",
@@ -363,6 +365,7 @@
363
  "floor": [
364
  "answer_decision",
365
  "ask_user",
 
366
  "extend_worker_budget",
367
  "pack_board",
368
  "recall",
@@ -385,6 +388,7 @@
385
  "floor": [
386
  "answer_decision",
387
  "ask_user",
 
388
  "extend_worker_budget",
389
  "fetch_url",
390
  "list_dir",
@@ -476,6 +480,7 @@
476
  ],
477
  "loadable": [
478
  "answer_decision",
 
479
  "diff",
480
  "extend_worker_budget",
481
  "git_blame",
@@ -673,6 +678,7 @@
673
  "floor": [
674
  "answer_decision",
675
  "ask_user",
 
676
  "delegate_dispatch",
677
  "edit",
678
  "extend_worker_budget",
@@ -714,6 +720,7 @@
714
  "floor": [
715
  "answer_decision",
716
  "ask_user",
 
717
  "edit",
718
  "extend_worker_budget",
719
  "fetch_url",
@@ -762,6 +769,7 @@
762
  "floor": [
763
  "answer_decision",
764
  "ask_user",
 
765
  "edit",
766
  "extend_worker_budget",
767
  "grep",
@@ -870,6 +878,7 @@
870
  "floor": [
871
  "answer_decision",
872
  "ask_user",
 
873
  "extend_worker_budget",
874
  "grep",
875
  "list_dir",
@@ -907,6 +916,7 @@
907
  "floor": [
908
  "answer_decision",
909
  "ask_user",
 
910
  "extend_worker_budget",
911
  "list_dir",
912
  "pack_board",
@@ -1017,6 +1027,12 @@
1017
  "option": "Copy files within write scope (files only; never use command cp).",
1018
  "card": "Tool copy: Copy files within write scope (files only; never use command cp)."
1019
  },
 
 
 
 
 
 
1020
  {
1021
  "name": "delegate_decompose",
1022
  "description": "Decompose an approved plan into delegation legs",
@@ -1031,15 +1047,15 @@
1031
  },
1032
  {
1033
  "name": "delegate_init",
1034
- "description": "Create a delegation departure for the current coordinator session",
1035
- "option": "Create a delegation departure for the current coordinator session",
1036
- "card": "Tool delegate_init: Create a delegation departure for the current coordinator session"
1037
  },
1038
  {
1039
  "name": "delegate_status",
1040
- "description": "Return delegation departure status and legs",
1041
- "option": "Return delegation departure status and legs",
1042
- "card": "Tool delegate_status: Return delegation departure status and legs"
1043
  },
1044
  {
1045
  "name": "delete",
@@ -1391,9 +1407,9 @@
1391
  },
1392
  {
1393
  "name": "request_budget",
1394
- "description": "Ask your coordinator for more tool rounds when this leg's goal needs more than remain. Name the rounds and the concrete work they cover. You keep working while the coordinator decides; a granted ceiling arrives as a host notice on your next round. If none arrives, finish with an honest partial report at your current ceiling. One request can be",
1395
- "option": "Ask your coordinator for more tool rounds when this leg's goal needs more than remain. Name the rounds and the concrete work they cover. You keep working while the coordinator decides; a granted ceiling arrives as a host notice on your next round. If none arrives, finish with an honest partial report at your current ceiling. One request can be",
1396
- "card": "Tool request_budget: Ask your coordinator for more tool rounds when this leg's goal needs more than remain. Name the rounds and the concrete work they cover. You keep working while the coordinator decides; a granted ceiling arrives as a host notice on your next round. If none arrives, finish with an honest partial report at your current ceiling. One request can be"
1397
  },
1398
  {
1399
  "name": "request_decision",
@@ -1403,9 +1419,9 @@
1403
  },
1404
  {
1405
  "name": "request_tools",
1406
- "description": "Load more tool schemas for the next model call by describing what the work needs",
1407
- "option": "Load more tool schemas for the next model call by describing what the work needs",
1408
- "card": "Tool request_tools: Load more tool schemas for the next model call by describing what the work needs"
1409
  },
1410
  {
1411
  "name": "restore_version",
@@ -1463,9 +1479,9 @@
1463
  },
1464
  {
1465
  "name": "skills_read",
1466
- "description": "Read a skill or resource by name, find skills by describing the work, or omit arguments to list every loaded skill.",
1467
- "option": "Read a skill or resource by name, find skills by describing the work, or omit arguments to list every loaded skill.",
1468
- "card": "Tool skills_read: Read a skill or resource by name, find skills by describing the work, or omit arguments to list every loaded skill."
1469
  },
1470
  {
1471
  "name": "source_history",
@@ -1668,6 +1684,15 @@
1668
  "stock": true,
1669
  "description": "Choosing the first read when a request is about an unfamiliar area, a named file or package, a JSON or YAML config, or the findings of a completed scan; the ladder from a repository map to a bounded read.",
1670
  "slot": "orientation",
 
 
 
 
 
 
 
 
 
1671
  "hosts": [
1672
  "coordinator",
1673
  "worker"
@@ -1939,6 +1964,19 @@
1939
  "stock": true,
1940
  "description": "Making a claim about how code behaves, how components connect, how a page or terminal renders, or what an image shows, and the receipt each kind of claim requires.",
1941
  "slot": "evidence",
 
 
 
 
 
 
 
 
 
 
 
 
 
1942
  "hosts": [
1943
  "coordinator",
1944
  "worker"
@@ -2327,8 +2365,8 @@
2327
  },
2328
  {
2329
  "name": "work-with-containers",
2330
- "description": "Build, debug, network, test, and publish containers; manage Compose, mounts, resources, and shutdown.",
2331
- "card": "Skill: work-with-containers\nBuild, debug, network, test, and publish containers; manage Compose, mounts, resources, and shutdown."
2332
  },
2333
  {
2334
  "name": "work-with-github",
 
6
  "head_tokens": 512
7
  },
8
  "questions": {
9
+ "deadline_ms": 5000,
10
  "tools": {
11
+ "independent": true,
12
  "question": "Which of these tools will completing this request call?",
13
  "option_words": 60,
14
+ "load_at": 0.94
15
+ },
16
+ "skills": {
17
+ "preload_at": 3.4
18
  },
19
  "guides": {
20
+ "omittable": [],
21
  "question": "Which of this guidance does completing this request need?",
22
  "option_words": 8,
23
+ "omit_below": 0,
24
  "confidence_floor": 0.5
25
  },
26
  "kind": {
 
34
  "inspect": "reading and understanding the project without changing it",
35
  "run": "building, testing, running, or verifying something"
36
  }
 
 
 
 
 
37
  }
38
  },
39
  "request": {
40
+ "deadline_ms": 5000,
41
+ "load_at": 1.95,
42
+ "max_loads": 4,
43
+ "nearest_loads": 1
44
  },
45
  "lookup": {
46
+ "deadline_ms": 5000,
47
+ "read_at": 3.2
 
48
  },
49
  "kinds": [
50
  "answer_only",
 
114
  "floor": [
115
  "answer_decision",
116
  "ask_user",
117
+ "decline_worker_budget",
118
  "edit",
119
  "extend_worker_budget",
120
  "grep",
 
160
  "floor": [
161
  "answer_decision",
162
  "ask_user",
163
+ "decline_worker_budget",
164
  "extend_worker_budget",
165
  "promote_overlay",
166
  "recall",
 
205
  "command_output",
206
  "command_stop",
207
  "copy",
208
+ "decline_worker_budget",
209
  "delete",
210
  "diff",
211
  "edit",
 
289
  "command",
290
  "command_output",
291
  "command_stop",
292
+ "decline_worker_budget",
293
  "extend_worker_budget",
294
  "fetch_url",
295
  "grep",
 
365
  "floor": [
366
  "answer_decision",
367
  "ask_user",
368
+ "decline_worker_budget",
369
  "extend_worker_budget",
370
  "pack_board",
371
  "recall",
 
388
  "floor": [
389
  "answer_decision",
390
  "ask_user",
391
+ "decline_worker_budget",
392
  "extend_worker_budget",
393
  "fetch_url",
394
  "list_dir",
 
480
  ],
481
  "loadable": [
482
  "answer_decision",
483
+ "decline_worker_budget",
484
  "diff",
485
  "extend_worker_budget",
486
  "git_blame",
 
678
  "floor": [
679
  "answer_decision",
680
  "ask_user",
681
+ "decline_worker_budget",
682
  "delegate_dispatch",
683
  "edit",
684
  "extend_worker_budget",
 
720
  "floor": [
721
  "answer_decision",
722
  "ask_user",
723
+ "decline_worker_budget",
724
  "edit",
725
  "extend_worker_budget",
726
  "fetch_url",
 
769
  "floor": [
770
  "answer_decision",
771
  "ask_user",
772
+ "decline_worker_budget",
773
  "edit",
774
  "extend_worker_budget",
775
  "grep",
 
878
  "floor": [
879
  "answer_decision",
880
  "ask_user",
881
+ "decline_worker_budget",
882
  "extend_worker_budget",
883
  "grep",
884
  "list_dir",
 
916
  "floor": [
917
  "answer_decision",
918
  "ask_user",
919
+ "decline_worker_budget",
920
  "extend_worker_budget",
921
  "list_dir",
922
  "pack_board",
 
1027
  "option": "Copy files within write scope (files only; never use command cp).",
1028
  "card": "Tool copy: Copy files within write scope (files only; never use command cp)."
1029
  },
1030
+ {
1031
+ "name": "decline_worker_budget",
1032
+ "description": "Decline a running worker's open request_budget. Its ceiling stays where it is; the worker is told on its next round to finish within it with an honest partial report, and its request closes.",
1033
+ "option": "Decline a running worker's open request_budget. Its ceiling stays where it is; the worker is told on its next round to finish within it with an honest partial report, and its request closes.",
1034
+ "card": "Tool decline_worker_budget: Decline a running worker's open request_budget. Its ceiling stays where it is; the worker is told on its next round to finish within it with an honest partial report, and its request closes."
1035
+ },
1036
  {
1037
  "name": "delegate_decompose",
1038
  "description": "Decompose an approved plan into delegation legs",
 
1047
  },
1048
  {
1049
  "name": "delegate_init",
1050
+ "description": "Create a delegation for the current coordinator session",
1051
+ "option": "Create a delegation for the current coordinator session",
1052
+ "card": "Tool delegate_init: Create a delegation for the current coordinator session"
1053
  },
1054
  {
1055
  "name": "delegate_status",
1056
+ "description": "Return delegation status and legs",
1057
+ "option": "Return delegation status and legs",
1058
+ "card": "Tool delegate_status: Return delegation status and legs"
1059
  },
1060
  {
1061
  "name": "delete",
 
1407
  },
1408
  {
1409
  "name": "request_budget",
1410
+ "description": "Ask your coordinator for more tool rounds as soon as you can see this leg's goal needs more than remain. Name the rounds and the concrete work they cover. You keep working while the coordinator decides; its answer arrives as a host notice on a later round, and the host holds your final round for that answer. If it declines",
1411
+ "option": "Ask your coordinator for more tool rounds as soon as you can see this leg's goal needs more than remain. Name the rounds and the concrete work they cover. You keep working while the coordinator decides; its answer arrives as a host notice on a later round, and the host holds your final round for that answer. If it declines",
1412
+ "card": "Tool request_budget: Ask your coordinator for more tool rounds as soon as you can see this leg's goal needs more than remain. Name the rounds and the concrete work they cover. You keep working while the coordinator decides; its answer arrives as a host notice on a later round, and the host holds your final round for that answer. If it declines"
1413
  },
1414
  {
1415
  "name": "request_decision",
 
1419
  },
1420
  {
1421
  "name": "request_tools",
1422
+ "description": "Load tools for the next concrete operation",
1423
+ "option": "Load tools for the next concrete operation",
1424
+ "card": "Tool request_tools: Load tools for the next concrete operation"
1425
  },
1426
  {
1427
  "name": "restore_version",
 
1479
  },
1480
  {
1481
  "name": "skills_read",
1482
+ "description": "Read the most relevant skill for a described need, or a referenced file from that skill.",
1483
+ "option": "Read the most relevant skill for a described need, or a referenced file from that skill.",
1484
+ "card": "Tool skills_read: Read the most relevant skill for a described need, or a referenced file from that skill."
1485
  },
1486
  {
1487
  "name": "source_history",
 
1684
  "stock": true,
1685
  "description": "Choosing the first read when a request is about an unfamiliar area, a named file or package, a JSON or YAML config, or the findings of a completed scan; the ladder from a repository map to a bounded read.",
1686
  "slot": "orientation",
1687
+ "needed_with": [
1688
+ "list_dir",
1689
+ "summarize",
1690
+ "survey_repo",
1691
+ "find",
1692
+ "jq",
1693
+ "scan_query",
1694
+ "scan_summary"
1695
+ ],
1696
  "hosts": [
1697
  "coordinator",
1698
  "worker"
 
1964
  "stock": true,
1965
  "description": "Making a claim about how code behaves, how components connect, how a page or terminal renders, or what an image shows, and the receipt each kind of claim requires.",
1966
  "slot": "evidence",
1967
+ "needed_with": [
1968
+ "capture_page",
1969
+ "page_snapshot",
1970
+ "page_act",
1971
+ "measure_page",
1972
+ "render_view",
1973
+ "view_image",
1974
+ "view_video",
1975
+ "terminal_snapshot",
1976
+ "verify",
1977
+ "web_search",
1978
+ "fetch_url"
1979
+ ],
1980
  "hosts": [
1981
  "coordinator",
1982
  "worker"
 
2365
  },
2366
  {
2367
  "name": "work-with-containers",
2368
+ "description": "Containerize apps with a Dockerfile; build Docker images; run, debug, network, test, and publish containers; manage Compose, mounts, resources, and shutdown.",
2369
+ "card": "Skill: work-with-containers\nContainerize apps with a Dockerfile; build Docker images; run, debug, network, test, and publish containers; manage Compose, mounts, resources, and shutdown."
2370
  },
2371
  {
2372
  "name": "work-with-github",
corpus.json CHANGED
The diff for this file is too large to render. See raw diff
 
holdout.jsonl CHANGED
The diff for this file is too large to render. See raw diff
 
train.jsonl CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ec3a0008eb903cd455912ce094a1e53884498a77ef76a20fa72805ab7b9aaeb5
3
- size 27801342
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12c4b2d46ee1088882be996f409b661c2b9741c5e97cd6a71e3d580102633685
3
+ size 27820718
training/turn-load/guide_audit.py ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Audit omissions including unlabeled units, and propose a supported allowlist.
2
+
3
+ Usage: guide_audit.py TRAIN_JSONL PREDICTIONS_JSONL OUT_JSON
4
+ Uses the installed B2 thresholds (.38 probability, .5 confidence). A unit must
5
+ have both label classes in training and validation, at least five examples of
6
+ each on validation, and at least five omissions at 97% precision with 98% needed-guide retention
7
+ to qualify.
8
+ Labels are observational proxies; this does not establish instruction usefulness.
9
+ """
10
+ import collections
11
+ import json
12
+ import sys
13
+ from pathlib import Path
14
+
15
+
16
+ def audit(train, predictions, threshold=.38, confidence=.5):
17
+ training = collections.defaultdict(collections.Counter)
18
+ for row in train:
19
+ for name, value in row['labels']['guides'].items():
20
+ if value is not None:
21
+ training[name]['positive' if value else 'negative'] += 1
22
+ stats = collections.defaultdict(collections.Counter)
23
+ for row in predictions:
24
+ for name, value in row['answers'].get('guides', {}).get('probabilities', {}).items():
25
+ needed = row['labels']['guides'].get(name)
26
+ omitted = value < threshold and max(value, 1 - value) >= confidence
27
+ stats[name]['scored'] += 1
28
+ stats[name]['omitted'] += omitted
29
+ if needed is None:
30
+ stats[name]['unknown'] += 1
31
+ stats[name]['omitted_unknown'] += omitted
32
+ else:
33
+ stats[name]['positive' if needed else 'negative'] += 1
34
+ stats[name]['omitted_needed' if needed else 'omitted_unneeded'] += omitted
35
+ allowed = []
36
+ for name, s in stats.items():
37
+ known_omissions = s['omitted_needed'] + s['omitted_unneeded']
38
+ if (training[name]['positive'] > 0 and training[name]['negative'] > 0
39
+ and s['positive'] >= 5 and s['negative'] >= 5 and not s['omitted_unknown']
40
+ and known_omissions >= 5 and s['omitted_unneeded'] / known_omissions >= .97
41
+ and 1 - s['omitted_needed'] / s['positive'] >= .98):
42
+ allowed.append(name)
43
+ return {'omittable': sorted(allowed), 'omit_below': threshold, 'confidence_floor': confidence,
44
+ 'training': {k: dict(v) for k, v in sorted(training.items())},
45
+ 'validation': {k: dict(v) for k, v in sorted(stats.items())},
46
+ 'limitation': 'Observed tool use supplies guide labels; unlabeled guidance cannot be certified for omission.'}
47
+
48
+
49
+ def main():
50
+ train, predictions, output = sys.argv[1:]
51
+ def read(path):
52
+ return [json.loads(s) for s in Path(path).read_text().splitlines() if s.strip()]
53
+
54
+ training, scored = read(train), read(predictions)
55
+ candidates = [audit(training, scored, step / 100) for step in range(1, 39)]
56
+ result = max(candidates, key=lambda r: sum(r['validation'][n]['omitted'] for n in r['omittable']))
57
+ if not result['omittable']:
58
+ result['omit_below'] = 0
59
+ result['original_threshold_audit'] = audit(training, scored)
60
+ result['selection_rule'] = 'Maximize supported omissions on validation; per-unit precision >= .97 and needed retention >= .98.'
61
+ Path(output).write_text(json.dumps(result, indent=2) + '\n')
62
+ print(json.dumps(result, indent=2))
63
+
64
+
65
+ if __name__ == '__main__':
66
+ main()
training/turn-load/relabel_guides.py ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Relabel guide units in archived rows from a corpus whose units declare `needed_with`.
2
+
3
+ A row's guide labels come from the tools its turn called: a unit was needed when the
4
+ turn called one of the tools it attaches to or one it is needed with. Archived rows keep
5
+ only the loadable tools a turn called (`labels.tools`), but every floor tool a unit
6
+ attaches to is already recorded through that unit's own label, so the turn's call set is
7
+ reconstructed exactly: the loadable calls plus the attached tools of every unit labelled
8
+ true. Units the corpus does not know keep their archived label.
9
+
10
+ Usage: relabel_guides.py CORPUS ROWS OUT
11
+ """
12
+ import argparse
13
+ import json
14
+ from pathlib import Path
15
+
16
+
17
+ def called_tools(row, units):
18
+ """The tools a row's turn called, as far as its labels show."""
19
+ called = set(row["labels"].get("tools") or [])
20
+ for uid, needed in (row["labels"].get("guides") or {}).items():
21
+ if needed and uid in units:
22
+ called.update(units[uid].get("attaches") or [])
23
+ return called
24
+
25
+
26
+ def relabel(row, units):
27
+ guides = dict(row["labels"].get("guides") or {})
28
+ if row.get("partial") or not guides:
29
+ return row, 0
30
+ called = called_tools(row, units)
31
+ changed = 0
32
+ for uid in guides:
33
+ unit = units.get(uid)
34
+ if unit is None:
35
+ continue
36
+ attaches = unit.get("attaches") or []
37
+ needed_with = unit.get("needed_with") or []
38
+ if not attaches and not needed_with:
39
+ continue
40
+ label = any(tool in called for tool in attaches) or any(tool in called for tool in needed_with)
41
+ if guides[uid] != label:
42
+ changed += 1
43
+ guides[uid] = label
44
+ row["labels"]["guides"] = guides
45
+ return row, changed
46
+
47
+
48
+ def main():
49
+ ap = argparse.ArgumentParser(description=__doc__)
50
+ ap.add_argument("corpus")
51
+ ap.add_argument("rows")
52
+ ap.add_argument("out")
53
+ args = ap.parse_args()
54
+ corpus = json.loads(Path(args.corpus).read_text())
55
+ units = {u["id"]: u for u in corpus["units"]}
56
+ counts = {}
57
+ total = changed_rows = 0
58
+ with Path(args.rows).open() as src, Path(args.out).open("x") as dst:
59
+ for line in src:
60
+ if not line.strip():
61
+ continue
62
+ row, changed = relabel(json.loads(line), units)
63
+ total += 1
64
+ changed_rows += bool(changed)
65
+ for uid, label in (row["labels"].get("guides") or {}).items():
66
+ key = (uid, "none" if label is None else ("true" if label else "false"))
67
+ counts[key] = counts.get(key, 0) + 1
68
+ dst.write(json.dumps(row) + "\n")
69
+ print("rows %d, rows with a changed label %d" % (total, changed_rows))
70
+ for uid in sorted({uid for uid, _ in counts}):
71
+ print(" %-28s true=%4d false=%4d none=%4d" % (uid, counts.get((uid, "true"), 0), counts.get((uid, "false"), 0), counts.get((uid, "none"), 0)))
72
+
73
+
74
+ if __name__ == "__main__":
75
+ main()
training/turn-load/scripts/decide/README.md ADDED
@@ -0,0 +1,358 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Turn decisions: engine, data, and evaluation
2
+
3
+ The why and the map are in [`docs/decision-engine.md`](../../docs/decision-engine.md); this is the training loop.
4
+
5
+ At every visible user turn, and at the start of every worker leg, the host
6
+ asks a local decision model which units the request needs: which loadable
7
+ tool schemas join the call and which instruction units stay out of the prompt.
8
+ Skills are ranked when the coordinator requests one through free text. The questions live in
9
+ `lycaon/config/packs/painted-wolf/platform/host/decisions.yaml`; the host
10
+ side is `lycaon/internal/decide` (engine client), `lycaon/internal/promptunit`
11
+ (the unit catalog), and `lycaon/internal/coordinator/turnload` (the decision
12
+ and the ledger). This directory holds the engine daemon and the offline
13
+ tooling that trains and scores it. Everything here is a bespoke experiment:
14
+ it runs outside the verification queue and never during automated tests.
15
+
16
+ ## What is decided, and what happens without an engine
17
+
18
+ | Unit | Decision | Without an answer |
19
+ |---|---|---|
20
+ | Loadable tool schema | one logical `multi` question, with each released tool option encoded independently; a tool loads when its P reaches `turn.tools.load_at` | only the floor is offered; `request_tools` loads the rest |
21
+ | Instruction unit (`shared/units/*.md`) | one `multi` question per turn, every unit an option; a unit is omitted when its P is below `omit_below` and its certainty reaches `confidence_floor` | every unit renders |
22
+ | Turn kind | one `choice` question; a confident `answer_only` vetoes tool loading | no veto |
23
+ | `skills_read` text | loaded skills ranked against the text; the best relevant skill is read | word overlap |
24
+ | `request_tools` text | loadable schemas ranked against the text | exact names and word overlap |
25
+
26
+ A joint turn head uses three engine rows: the state is encoded once per row, and
27
+ each option is a `[MASK]` marker the head reads on its own. A yes/no
28
+ question per option would put the state through the encoder once per option,
29
+ about 77 times per turn; the three-row set costs 0.25 s p50 on MLX. Option
30
+ texts are the tool name plus `option_words` words of its description,
31
+ exported into the corpus so training reads exactly what the host asks. `state.head_tokens` is the
32
+ engine's budget for the options; the request text gets what remains.
33
+
34
+ Each kind defaults to the behaviour the surface has without a model, and the
35
+ engine only moves it toward the failure the model can recover from: a
36
+ missing tool costs a `request_tools` round trip, an extra instruction costs
37
+ bytes. A unit attached to loadable tools follows them and is never scored.
38
+
39
+ Switched off (`LYCAON_DECIDE_DISABLED=1`), without a binary or checkpoint, or
40
+ past its deadline, the engine abstains and every site keeps the surface's own
41
+ default: the floor is offered, every loadable tool is listed as requestable
42
+ with the same bounded description the engine would have scored, every
43
+ instruction unit renders, every skill lists, and nothing is pre-read. That
44
+ is the only difference between off and on: on, the head adds likely tools to
45
+ the call and omits guides it is confident about; it never removes a tool the
46
+ floor offers, and the kind veto ships off (`kind.veto_tools`).
47
+
48
+ ## Engine
49
+
50
+ The host launches Painted Wolf Decide (`pw-decide`,
51
+ `lycaon/internal/decide/native`) as a stdio subprocess speaking line-delimited JSON (`hello`,
52
+ `decide`, `rank`). One backbone stays resident and every request names the
53
+ head it wants; a head is a small set of weights swapped over the shared
54
+ encoder, so turn decisions and ranking share one memory footprint and one
55
+ forward path:
56
+
57
+ | Head | Trained by | Serves |
58
+ |---|---|---|
59
+ | `turn-load` | `scripts/decide/train.py` | tool and unit decisions, turn kind |
60
+ | `unit-rank` | `scripts/decide/train.py --families skills` | skill roster, `skills_read`, and `request_tools` ranking |
61
+ | `code-rank` | `scripts/decide/rerank/train_rerank.py` | summarize, repomap, and project-search reranking |
62
+ | `web-rank` | web-page pairs | verified web pages and search snippets |
63
+
64
+ A head is a safetensors file whose header names the backbone it was trained
65
+ over (`scripts/decide/headfile.py`); the engine refuses one trained over
66
+ another backbone, and a request for a head that did not load answers with the
67
+ checkpoint's own. The host resolves `pw-decide` beside its own executable and
68
+ the heads staged under `engine-root/decide/heads/<name>.safetensors`;
69
+ `LYCAON_DECIDE_HEADS="turn-load=PATH,unit-rank=PATH,code-rank=PATH"` names them outright
70
+ and `LYCAON_DECIDE_BINARY` / `LYCAON_DECIDE_MODEL_DIR` point a checkout at a
71
+ build and a checkpoint (`./task build:decide`, `pw decide ensure`). A packaged
72
+ app resolves the checkpoint `stage-engine.sh` bundles under
73
+ `engine-root/decide/models/`.
74
+ `LYCAON_DECIDE_DISABLED=1` switches the engine off; every decision then falls
75
+ back as the table above says.
76
+
77
+ On Apple silicon the engine runs the model on Apple's MLX (`--device mlx`,
78
+ what `auto` picks there): candle's Metal backend costs about five times more
79
+ per row, and the MLX port in `native/src/mlx.rs` mirrors the candle port op
80
+ for op, which its parity test checks on a tiny checkpoint. MLX's kernels
81
+ ship as `engine-root/decide/mlx.metallib` (about 136 MB, the framework's
82
+ whole kernel library); the host passes that path as `--metallib`, and the
83
+ engine also finds a copy beside its own executable. Building the engine on a
84
+ Mac compiles MLX from source once (CMake, network for the framework
85
+ checkout, and Xcode's Metal toolchain: `xcodebuild -downloadComponent
86
+ MetalToolchain` when the build says it cannot execute `metal`). The engine
87
+ caps MLX's allocator cache at 512 MB when it loads: every request has its own
88
+ batch size and sequence length, so an uncapped cache keeps growing toward the
89
+ GPU's working set (20 GB was measured on a rerank eval), and a resident
90
+ sidecar must not do that to the machine. With the cap the same eval holds
91
+ the engine at about 1.5 GB.
92
+
93
+ Heads are training artifacts. The shared release manifest,
94
+ `lycaon/config/packs/painted-wolf/platform/host/decision-release.json`, pins
95
+ their hashes, labels, backbone, and initial-preload option vocabulary. A
96
+ development checkout installs the matching set under
97
+ `<artifact bin>/decide-heads/<name>.safetensors`
98
+ (`python3 scripts/artifact_paths.py bin .` prints the directory; names are
99
+ `turn-load`, `unit-rank`, `code-rank`, `web-rank`), and every `build:lycaon-dev` stages
100
+ them into the engine root that `den:sidecar` and `den:app` run against.
101
+ Missing or mismatched required heads fail staging. The host intersects the
102
+ release vocabulary with permitted tools; new tools remain requestable without
103
+ changing the preloader's encoded options. Refresh the vocabulary and evaluate it
104
+ with its head before releasing it. The factory's `release_manifest.py` rebuilds
105
+ the manifest from the selected artifacts and their evaluated corpus.
106
+
107
+ Python is training-side only. `parity_probe.py --model ID --head FILE
108
+ --examples FILE --engine <launcher>` scores turns through the trainer's own
109
+ forward and through `pw-decide`. Report the actual probability gaps and
110
+ threshold crossings for the loaded heads; metadata compatibility alone does
111
+ not establish numerical parity, and agreement is not a model-quality test.
112
+ The training environment:
113
+
114
+ ```bash
115
+ python3 -m venv .venv-decide && .venv-decide/bin/pip install laya
116
+ ```
117
+
118
+ ## Backbone
119
+
120
+ Laya ships two backbones with different head shapes, so a head only fits the
121
+ backbone it was trained on:
122
+
123
+ | `LYCAON_DECIDE_MODEL_ID` | Backbone | Hidden | Context |
124
+ |---|---|---|---|
125
+ | `convaiinnovations/laya` | ModernBERT-large, English vocabulary | 1024 | 512 |
126
+ | `convaiinnovations/laya-multilingual` | mmBERT-base, 100+ languages | 768 | 1024 |
127
+
128
+ The backbone is one choice for every head: the heads share the encoder, so
129
+ the replay eval here and the rerank eval must agree on it before either
130
+ ships. The shipped backbone is `convaiinnovations/laya-multilingual`: on the
131
+ same turn data it scored no worse than the English checkpoint and ran the
132
+ turn set in 0.81 s against 1.32 s p50 on candle Metal. Changing it means
133
+ training the turn-load head and running the replay eval under each id on the
134
+ held-out captures.
135
+
136
+ ## Corpus
137
+
138
+ The host exports the corpus the decision scores; nothing here parses packs:
139
+
140
+ Artifacts live in the checkout's resolved build directory, never under the
141
+ checkout itself; every command below uses that directory as `$DECIDE`:
142
+
143
+ ```bash
144
+ DECIDE="$(python3 scripts/artifact_paths.py build .)/decide"
145
+ BUILD_ONLY=true ./task eval:tool-usage # builds lycaon-debug; or LYCAON_DEBUG_BINARY=... pointing at one
146
+ scripts/decide/corpus.py --out "$DECIDE/corpus.json"
147
+ ```
148
+
149
+ `corpus.json` carries every coordinator surface's floor, loadable tools, and
150
+ scored units; every tool's option text and request card; every unit and
151
+ skill card; the question templates; and the catalog revision that receipts
152
+ record. Third-party packs contribute units through the same catalog, so the
153
+ corpus is whatever the effective catalog resolves to.
154
+
155
+ ## Data
156
+
157
+ Every decided turn leaves a receipt, including a turn whose engine was off or
158
+ abstained, and `lycaon-debug decide export` is the one reader that turns
159
+ receipts into training rows (`pw-decide-row/1`, described by
160
+ [`row.schema.json`](row.schema.json) and loaded through `rows.py`). A row is
161
+ the state the engine read, verbatim; the candidates the turn offered
162
+ (`offered`); what the engine answered (`engine`); and what the session then
163
+ did (`labels`). Coordinator turns and worker legs export alike: a worker
164
+ leg's receipt names its tool profile as the surface and its coordinator
165
+ session as `root_session`.
166
+
167
+ ```bash
168
+ lycaon-debug decide export --db <store.db> --out rows.jsonl [--roots roots.txt]
169
+ ```
170
+
171
+ Labels are always the host's own vocabulary and come from what the session
172
+ actually did: the offered loadable tools it called, the ones it asked for by
173
+ name, each `request_tools` need with the names it spelled out and the tools
174
+ it went on to call, the skills it read, and the kind those calls imply
175
+ (`turnload.ObservedKind`). A scored instruction unit attached to tools is
176
+ needed exactly when one of them was called; an unattached unit has no
177
+ behavioural label (`turnload.GuideLabels`). A turn that did not complete
178
+ keeps only its needs.
179
+
180
+ A tool a session called is weak evidence that the request needed it: driving
181
+ models call tools out of habit, and a head trained on calls learns which
182
+ tools are common rather than which a request needs. When judges have scored
183
+ the turn's loadable tools against its request (`labels.tool_scores`, and a
184
+ second judge's `labels.second_scores`), a tool is needed when both judges
185
+ scored it likely or certain, unneeded when both scored it at most unlikely,
186
+ and unlabelled otherwise, so the head never trains on a call two judges
187
+ disagree about; a tool the model asked for by name is always needed. A card
188
+ only one judge scored is unlabelled. Skill and need levels are the two
189
+ judges' mean, except that by default (`--rank-levels blended`) the skill a
190
+ turn read first and the tools used after a need train at the top level
191
+ whatever the judges scored them; `--rank-levels skills-blended` keeps the
192
+ override for skill reads only, and `--rank-levels judged` drops it. `--tool-truth` on the trainer,
193
+ calibration, and replay chooses the rule (`rows.TOOL_TRUTH`); rows without
194
+ tool scores fall back to the called tools.
195
+
196
+ A row whose `engine.state` is `answered` carried what the engine chose, so a
197
+ tool the engine preloaded and the session then called is not an independent
198
+ label: the turn families train on the rows the engine did not answer, and
199
+ under a live engine a turn's needs are exactly the tools it missed, which is
200
+ what the rank head ranks in production.
201
+
202
+ Open training data comes from the dataset factory, `paintedwolf-decide`: it
203
+ drives sessions with open-weights models in sandboxed runners against pinned
204
+ public repositories (through `lycaon-debug decide generate`, which drives a
205
+ JSON-lines task file through one sidecar and records a manifest), exports
206
+ their receipts, has an open-weights judge of another family score skill and
207
+ tool cards 0..4, and splits by repository and prompt group. Rows from your
208
+ own stores export the same way and train the same way; they stay on the
209
+ machine that exported them.
210
+
211
+ Generation tasks that select a workflow provide both `workflow` and
212
+ `workflow_version`. The runner verifies that exact catalog identity before
213
+ creating task sessions and records it in each result manifest; it does not
214
+ select a version implicitly. Omit both fields for ordinary chat tasks.
215
+ For example, `{"id":"review","prompt":"Review this change","workflow":"implement","workflow_version":"1.0.0"}` pins that definition.
216
+
217
+ ## Train, calibrate, evaluate
218
+
219
+ Hold out whole packs and whole repositories: units from `--holdout-pack` stay
220
+ out of the guide labels so the eval measures generalization to units the
221
+ head never saw (what lets extensions ship their own units without
222
+ retraining), and the factory holds out every row of its held-out
223
+ repositories. The rest splits by prompt group, so the validation set is
224
+ prompts the head never trained on.
225
+
226
+ ```bash
227
+ # turn-load answers the turn questions; unit-rank ranks skill and tool cards.
228
+ # The rank pairs are kept out of turn-load because they outnumber its rows and
229
+ # pull the shared weights. On a GPU host, set LYCAON_DECIDE_DEVICE=cuda.
230
+ scripts/decide/train.py --corpus corpus.json --train train.jsonl --val val.jsonl --holdout-pack painted-wolf/browser \
231
+ --families tools,guides,kind --tool-weight sqrt-inverse --seed 11 --out heads/turn-load.safetensors
232
+ scripts/decide/train.py --corpus corpus.json --train train.jsonl --val val.jsonl \
233
+ --families skills,requests --skill-scored 4 --skill-zeros 3 --out heads/unit-rank.safetensors
234
+
235
+ # Replay through the shipped engine, calibrate the thresholds, and time the turn set.
236
+ # The launcher execs: pw-decide serve --model <checkpoint dir> --model-id <id> --device mlx
237
+ # --head-max-len <state.head_tokens> --head turn-load=<head> --head unit-rank=<head>
238
+ export LYCAON_DECIDE_ENGINE=<launcher>
239
+ scripts/decide/replay_eval.py --corpus corpus.json --examples holdout.jsonl --holdout-pack painted-wolf/browser --json replay-holdout.json
240
+ scripts/decide/calibrate.py --corpus corpus.json --examples val.jsonl
241
+ scripts/decide/bench_latency.py --corpus corpus.json
242
+ ```
243
+
244
+ The launcher runs the same binary with the same arguments the host uses, so
245
+ the replay measures the shipped engine; `parity_probe.py` cross-checks a
246
+ head against the trainer's own forward.
247
+
248
+ `train.py` precomputes the frozen backbone's features once and trains the
249
+ head for up to sixty epochs, stopping after twelve without a better
250
+ selection loss: the validation loss of tools and guides, the families that
251
+ change what a turn carries. Kind is trained and reported but does not pick
252
+ the checkpoint; its loss swings enough to stop a run before the tools head
253
+ has learned anything. The trainer reports precision and recall per family,
254
+ and per host for coordinator turns and worker legs, rather than one pooled
255
+ number, because the guide rows carry many always-true units that would hide
256
+ an unlearned tools head. Option sets come from each row's `offered`
257
+ candidates and option texts from the corpus, choice labels index options in
258
+ the engine's order (sorted by name), and skill pairs use the corpus's cards,
259
+ so the head sees at training exactly what it answers at the turn.
260
+ `--tool-weight sqrt-inverse` weighs each tool's positives by
261
+ sqrt(N / (n + 1)), clamped to [1, 20], so a tool few turns use still pulls
262
+ the head. Check a new head with the trainer's own forward before blaming the
263
+ engine: `pw-decide` matches it to the fourth decimal.
264
+
265
+ `replay_eval.py` reports tool load precision and recall (micro, and macro
266
+ over tools), loads per turn, guide omission precision and recall, kind
267
+ accuracy, skill top-1 and the share of turns whose first-read skill the
268
+ pruned roster listed, against judged skill scores the preload precision and
269
+ the share of turns whose roster lists a relevant skill, need ranking, bytes
270
+ saved per turn, the share of requests cut at `user_text_chars`, and engine
271
+ latency; overall and by host, surface, language, project, and the model that
272
+ drove the session, and for the held-out packs.
273
+ `calibrate.py` also reports the tool threshold each host would choose alone.
274
+ `bench_latency.py` measures the full turn question set on this machine,
275
+ which is the number the `deadline_ms` budget must respect.
276
+
277
+ A head ships when it beats the engine-off default on the held-out sets and
278
+ is at least as good as the shipped head on every set both can be scored on.
279
+ Off, no loadable tool is preloaded and every guide renders, so any tool
280
+ recall saves round trips, while a wrongly omitted guide costs the turn;
281
+ guide omission therefore turns on only when its precision on the held-out
282
+ repositories reaches 0.97. The rest are targets the shipped head reports
283
+ against: tool recall ≥ 0.9 (a missed tool costs a `request_tools` round
284
+ trip, so recall outranks precision), skill top-1 ≥ 0.7, first-read skill
285
+ listed ≥ 0.95, bytes saved ≥ 30% on investigate turns. On the generated
286
+ sessions a turn averages 11.8 coordinator calls of 7.5 s and 0.46
287
+ `request_tools` round trips, so the engine pays for itself once it avoids
288
+ about a quarter of them. The latency budget is the catalog's `deadline_ms`,
289
+ 5 s for turn, request, lookup, and tool-event decisions: the decision runs once before
290
+ the turn's first model call, and one avoided round trip pays for many seconds
291
+ of it, so accuracy is the binding bar, not the engine's wall time.
292
+
293
+ ## Author probe
294
+
295
+ `lycaon-debug decide probe --request "..." --surface implement_investigate`
296
+ runs the turn decision through the engine resolved from the environment and
297
+ prints what it would load and omit, so a pack author can see whether their
298
+ unit's description loads for the requests they intend.
299
+
300
+ ## Shipped heads: B5 tools, B7G guides, E4 skills, open1 code
301
+
302
+ The release manifest pins four heads over the open1 dataset: B5 `turn-load`
303
+ for tools, B7G `guide-load` for guides, E4-dense1 `unit-rank` for skills and
304
+ needs, and open1 `code-rank`. The host asks the guides question of the guide
305
+ head when the release ships one and of `turn-load` otherwise.
306
+
307
+ B5 encodes each tool independently: consensus labels, 24 inverse-weighted
308
+ sampled negatives per turn, no rare-tool weighting or augmentation, learning
309
+ rate 5e-4, seed 11, batch 64, a 45-epoch cap. On validation it scores
310
+ 0.822 precision and 0.291 recall at 0.95 with 0.308 preloads per turn; the
311
+ shipped cutoff is 0.85 (`decisions.yaml`), which on recorded sessions loads
312
+ five times as often at the same precision. Diagnostic requests outside
313
+ training show that initial prediction still misses useful tools; `request_tools`
314
+ covers the rest, with coverage 92.4% and top-1 77.7% at 2.59 loads per need
315
+ on the recorded validation check.
316
+
317
+ E4 uses two-judge labels and a seeded half-family augmentation mix. On 789
318
+ acceptance requests, 660 with a consensus-relevant skill, a relevant skill
319
+ was visible in the first six for 92.1%, and automatic preloads were relevant
320
+ in 122 of 123 cases; common-skill visibility is 87.9%, rare-skill 80.4%.
321
+
322
+ Stage the head set with `PW_DECIDE_HEADS_DIR` when building a checkout; the
323
+ build verifies it against the committed release manifest.
324
+ The committed release manifest binds its exact weights and option texts.
325
+
326
+ ## Guide omission: B7
327
+
328
+ B7 trains the B5 recipe with `--families tools,guides`: every tool and every
329
+ guide option on its own row. Guide labels are relabelled from tool calls
330
+ before training: a unit was needed when its turn called a tool it `attaches`
331
+ to or one it is `needed_with`. `claim-evidence` and `survey-first-pass` carry
332
+ labels for the first time; earlier releases left them unknown, so no audit
333
+ could certify them.
334
+
335
+ The host only omits ids listed in `turn.guides.omittable`; an empty list
336
+ retains every guide. `train-host/guide_audit.py` certifies a unit at 97%
337
+ omission precision and 98% needed-guide retention on validation and on the
338
+ held-out repositories, and never from an unknown label. Keep that report with
339
+ the installed head and thresholds; replays enforce the same list.
340
+
341
+ Guides-only control (B7G, `--families guides`, 21 epochs, patience stop),
342
+ scored with `train-host/turn_probe.py` on the relabelled rows:
343
+
344
+ | Unit | Holdout omissions | Unneeded | Needed retention |
345
+ |---|---|---|---|
346
+ | native-summarize-tool | 47 of 674 | 100% | 100% |
347
+ | native-find-tool | 6 of 674 | 100% | 100% |
348
+ | claim-evidence | 93 of 680 | 97.8% | 97.8% |
349
+ | native-recall-tool | 640 of 680 | 98.1% | 25% (16 positives) |
350
+ | survey-first-pass | 0 of 680 | | |
351
+
352
+ Certified at `omit_below` 0.15: `native-summarize-tool`, `native-find-tool`.
353
+ `claim-evidence` misses retention by 0.2 points on holdout and precision on
354
+ validation (92.8%); `native-recall-tool` omits almost always and misses the
355
+ few turns that called it. Labels are observational: a unit whose tool is
356
+ rarely called has few positives, so retention swings on a handful of turns.
357
+ The shipped list and any decision past the audit are recorded in
358
+ `decisions.yaml` beside the numbers.
training/turn-load/scripts/decide/corpus.py CHANGED
@@ -45,7 +45,7 @@ class Corpus:
45
  self.kinds = list(data.get("kinds") or KINDS)
46
  self.surfaces = {row["id"]: row for row in data["surfaces"]}
47
  self.tools = {row["name"]: row["description"] for row in data["tools"]}
48
- self.tool_options = {row["name"]: row.get("option", "") for row in data["tools"]}
49
  self.state_spec = data.get("state", {})
50
  self.units = {row["id"]: row for row in data["units"]}
51
  self.skills = {row["name"]: row for row in data.get("skills") or []}
@@ -78,18 +78,25 @@ class Corpus:
78
  else:
79
  options = {uid: self.units[uid].get("option", "") for uid in ids if uid in self.units}
80
  question = {"type": "multi", "instructions": spec["question"], "options": dict(sorted(options.items()))}
81
- if spec.get("independent"):
82
  question["independent"] = True
83
  return question
84
 
 
 
 
 
 
85
  def questions(self, loadable, guides):
86
- """The whole turn set the host asks over these candidates: tools, guides, kind."""
 
87
  qs = {}
88
  for kind, qid, ids in (("tool", "tools", loadable), ("guide", "guides", guides)):
89
  q = self.multi_question(kind, ids)
90
  if q["options"]:
91
  qs[qid] = q
92
- qs["kind"] = self.kind_question()
 
93
  return qs
94
 
95
  def row_questions(self, row):
 
45
  self.kinds = list(data.get("kinds") or KINDS)
46
  self.surfaces = {row["id"]: row for row in data["surfaces"]}
47
  self.tools = {row["name"]: row["description"] for row in data["tools"]}
48
+ self.tool_options = dict(self.spec["tools"].get("options") or {row["name"]: row.get("option", "") for row in data["tools"]})
49
  self.state_spec = data.get("state", {})
50
  self.units = {row["id"]: row for row in data["units"]}
51
  self.skills = {row["name"]: row for row in data.get("skills") or []}
 
78
  else:
79
  options = {uid: self.units[uid].get("option", "") for uid in ids if uid in self.units}
80
  question = {"type": "multi", "instructions": spec["question"], "options": dict(sorted(options.items()))}
81
+ if self.independent():
82
  question["independent"] = True
83
  return question
84
 
85
+ def independent(self):
86
+ """Whether the turn's multi questions encode one option per row. The tools
87
+ setting governs every multi question, so tools and guides share one encoding."""
88
+ return bool(self.spec["tools"].get("independent"))
89
+
90
  def questions(self, loadable, guides):
91
+ """The whole turn set the host asks over these candidates: tools and guides, and
92
+ the kind choice for a joint head (an independent head reads no roster)."""
93
  qs = {}
94
  for kind, qid, ids in (("tool", "tools", loadable), ("guide", "guides", guides)):
95
  q = self.multi_question(kind, ids)
96
  if q["options"]:
97
  qs[qid] = q
98
+ if not self.independent():
99
+ qs["kind"] = self.kind_question()
100
  return qs
101
 
102
  def row_questions(self, row):
training/turn-load/scripts/decide/parity_probe.py CHANGED
@@ -1,8 +1,5 @@
1
  #!/usr/bin/env python3
2
- """Score turns through the trainer's own forward and, optionally, through a pw-decide
3
- launcher, so a head can be checked before the engine is blamed: the two paths agree
4
- to the fourth decimal when the engine is right, and a head that scores nothing here
5
- was never trained, whatever its training log said.
6
 
7
  The head must record the encoding settings it trained with, and they must be the
8
  ones the engine serves at. Exit 0 when they match and the engine agrees within
@@ -36,19 +33,13 @@ def torch_probs(agent, model, corpus, ex, device):
36
  context = int(agent.cfg.get("max_len", 512))
37
  head_max_len = min(int(corpus.state_spec.get("head_tokens", 512)), context - 64)
38
  tools_q = corpus.multi_question("tool", ex["offered"]["loadable"])
39
- questions = [tools_q]
40
- if tools_q.get("independent"):
41
- questions = [dict(tools_q, options={k: v}) for k, v in tools_q["options"].items()]
42
- items = [multi_item(agent.tok, state_text(ex["state"]), q, {k: 0 for k in q["options"]}, context, head_max_len, "tools") for q in questions]
43
- features = precompute(agent, items, device)
44
- probabilities = {}
45
- for q, item, f in zip(questions, items, features):
46
- names = list(q["options"])[:len(item["markers"])]
47
- with torch.no_grad():
48
- logits = forward_head(model, f["h"][None].to(device), f["att"][None].to(device), f["marker_pos"][None].to(device),
49
- f["marker_mask"][None].to(device), torch.tensor([item["qtype"]], device=device))[0].cpu()
50
- probabilities.update(zip(names, torch.sigmoid(logits[:len(names)]).tolist()))
51
- return probabilities
52
 
53
 
54
  def engine_probs(proc, corpus, ex):
@@ -68,7 +59,7 @@ def main():
68
  ap.add_argument("--n", type=int, default=6)
69
  ap.add_argument("--tolerance", type=float, default=1e-3, help="the largest engine-trainer probability gap that passes")
70
  args = ap.parse_args()
71
- device = os.environ.get("LAYA_DEVICE") or ("mps" if torch.backends.mps.is_available() else "cpu")
72
  agent = laya.load(args.model, device=device)
73
  model = agent.model
74
  tensors = load_file(args.head, device=device)
@@ -90,10 +81,6 @@ def main():
90
  meta["max_len"], meta["head_max_len"], context, budget))
91
  return 1
92
  corpus = Corpus.load(args.corpus)
93
- encoding = "independent" if corpus.spec["tools"].get("independent") else "joint"
94
- if meta.get("tool_encoding", "joint") != encoding:
95
- print("parity: head tool encoding %s differs from corpus %s" % (meta.get("tool_encoding", "joint"), encoding))
96
- return 1
97
  proc = None
98
  if args.engine:
99
  proc = subprocess.Popen([args.engine], stdin=subprocess.PIPE, stdout=subprocess.PIPE, text=True, bufsize=1)
 
1
  #!/usr/bin/env python3
2
+ """Compare trainer and native-engine predictions on identical turn inputs.
 
 
 
3
 
4
  The head must record the encoding settings it trained with, and they must be the
5
  ones the engine serves at. Exit 0 when they match and the engine agrees within
 
33
  context = int(agent.cfg.get("max_len", 512))
34
  head_max_len = min(int(corpus.state_spec.get("head_tokens", 512)), context - 64)
35
  tools_q = corpus.multi_question("tool", ex["offered"]["loadable"])
36
+ item = multi_item(agent.tok, state_text(ex["state"]), tools_q, {k: 0 for k in tools_q["options"]}, context, head_max_len, "tools")
37
+ names = list(tools_q["options"].keys())[: len(item["markers"])]
38
+ f = precompute(agent, [item], device)[0]
39
+ with torch.no_grad():
40
+ logits = forward_head(model, f["h"][None].to(device), f["att"][None].to(device), f["marker_pos"][None].to(device),
41
+ f["marker_mask"][None].to(device), torch.tensor([item["qtype"]], device=device))[0].cpu()
42
+ return dict(zip(names, torch.sigmoid(logits[: len(names)]).tolist()))
 
 
 
 
 
 
43
 
44
 
45
  def engine_probs(proc, corpus, ex):
 
59
  ap.add_argument("--n", type=int, default=6)
60
  ap.add_argument("--tolerance", type=float, default=1e-3, help="the largest engine-trainer probability gap that passes")
61
  args = ap.parse_args()
62
+ device = os.environ.get("LYCAON_DECIDE_DEVICE") or ("mps" if torch.backends.mps.is_available() else "cpu")
63
  agent = laya.load(args.model, device=device)
64
  model = agent.model
65
  tensors = load_file(args.head, device=device)
 
81
  meta["max_len"], meta["head_max_len"], context, budget))
82
  return 1
83
  corpus = Corpus.load(args.corpus)
 
 
 
 
84
  proc = None
85
  if args.engine:
86
  proc = subprocess.Popen([args.engine], stdin=subprocess.PIPE, stdout=subprocess.PIPE, text=True, bufsize=1)
training/turn-load/scripts/decide/rerank/README.md ADDED
@@ -0,0 +1,173 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Reranking with the decision engine: data, training, evaluation
2
+
3
+ The host ranks candidates against a task at several seams: the files and
4
+ definitions a `summarize` pack admits, the symbols on a repomap page, the hits
5
+ Den project search returns, the pages web research verifies. Each seam keeps
6
+ its own lexical order and, when its site is enabled in
7
+ `lycaon/config/packs/painted-wolf/platform/host/decisions.yaml`, blends in the
8
+ decision engine's relevance for the lexical top K. The seam is
9
+ `lycaon/internal/decide` (`Reranker`); the engine is `pw-decide`
10
+ (`lycaon/internal/decide/native`), the Painted Wolf Decide sidecar, and `lycaon/internal/decide/pwdecide` is its client.
11
+
12
+ This directory holds the offline tooling that answers one question per site:
13
+ does the engine beat the lexical order, by how much, and at what latency.
14
+ Everything here is a bespoke experiment: it runs outside the verification
15
+ queue and never during automated tests. Training needs Python; serving does
16
+ not.
17
+
18
+ ## Engine
19
+
20
+ ```bash
21
+ ./task build:decide # <build dir>/pw-decide, Metal on Apple silicon
22
+ pw-decide check --model <checkpoint dir> --head code-rank=heads/code-rank.safetensors
23
+ pw-decide bench --model <checkpoint dir> --candidates 16
24
+ ```
25
+
26
+ The checkpoint directory is a Laya checkpoint (`rl_agent_config.json`,
27
+ `model.safetensors`, `encoder/config.json`, `tokenizer/`); the host provisions
28
+ the pinned one with `pw decide ensure`, and a Hugging Face cache snapshot has
29
+ the same layout. One backbone stays resident and every request names its head
30
+ (`turn-load`, `code-rank`, `web-rank`); a head that did not load answers with
31
+ the checkpoint's own. The engine refuses a head trained over another backbone.
32
+
33
+ The evaluation command runs the host's own client, so it needs the same
34
+ environment the host would resolve:
35
+
36
+ ```bash
37
+ export LYCAON_DECIDE_BINARY="$(python3 scripts/artifact_paths.py build "$PWD")/pw-decide"
38
+ export LYCAON_DECIDE_MODEL_DIR=~/.config/paintedwolf-dev/decide-models/convaiinnovations--laya-multilingual@e4e9ddf21a7b
39
+ export LYCAON_DECIDE_HEADS="code-rank=$PWD/.task/decide/heads/code-rank.safetensors"
40
+ ```
41
+
42
+ ## Corpus and pairs
43
+
44
+ `decide-rerank` is the evaluation command. Build it once into the ignored
45
+ task directory:
46
+
47
+ ```bash
48
+ go build -o .task/decide/bin/decide-rerank ./lycaon/cmd/decide-rerank
49
+ ```
50
+
51
+ Harvest definitions with the host's own parsers, then write pairs. A pair is
52
+ a request whose answer is one unit. Pairs come from the dataset factory
53
+ (`pwdecide coderank`, in paintedwolf-decide), which keeps each unit's leading
54
+ comment as a free, human-written request and has pinned open-weights models
55
+ write requests across a language and register panel, so every training
56
+ request is reproducible from open models. A doc-derived request is the
57
+ candidate's own text once the candidate carries its leading comment, so only
58
+ the model-written requests measure anything on doc-bearing text.
59
+
60
+ ```bash
61
+ B=.task/decide/bin/decide-rerank; D=.task/decide/data
62
+ $B harvest --repo lycaon --name lycaon --include internal --out $D/units-lycaon.jsonl
63
+ # In paintedwolf-decide: pwdecide coderank pairs --units <units dir> --out <pairs dir>
64
+ scripts/decide/rerank/synthesize_sites.py code --units $D/units-lycaon.jsonl --pairs $D/pairs/model-lycaon.jsonl \
65
+ --repo lycaon --out $D/rows-sites-lycaon.jsonl
66
+ ```
67
+
68
+ `synthesize_sites.py` writes rows for the site shapes the harvest cannot
69
+ produce (search hits, neighbours, imports, next actions, web pages) with
70
+ explicit rubric labels. Held-out repositories never contribute training rows.
71
+
72
+ ## Evaluate a site
73
+
74
+ `eval` drives the site through its real entry point (the summarize engine,
75
+ `repomap.Build`, the live code leg) with the engine attached, and reports the
76
+ target's rank under the site's lexical order and under the blend:
77
+
78
+ ```bash
79
+ $B eval --site summarize_definitions --repo ../ast-grep --name ast-grep \
80
+ --units $D/units-ast-grep.jsonl --pairs $D/pairs/model-ast-grep.jsonl --k 16 --chunk 16 --deadline 8s \
81
+ --json .task/decide/eval/definitions-ast-grep.json --dump $D/dump-definitions-ast-grep.jsonl
82
+ ```
83
+
84
+ `--no-engine` runs the lexical order only and still writes the dump, which is
85
+ how training rows are produced without spending engine time. `--weight`,
86
+ `--k`, `--chunk`, and `--deadline` (a duration) override the catalog policy
87
+ for a sweep; measure with a generous deadline, because a call the engine loses
88
+ to the deadline counts as unchanged. Run one engine at a time on a laptop GPU:
89
+ two engines sharing it both miss their deadlines.
90
+
91
+ The report gives MRR, nDCG@10, and hit@1/5/10 for lexical and blended, how
92
+ many pairs improved or regressed, abstention reasons, and p50/p95 latency of
93
+ the engine call. `web_pages` has no offline corpus and is measured live.
94
+
95
+ ## Train
96
+
97
+ ```bash
98
+ scripts/decide/rerank/train_rerank.py --dump $D/dump-definitions-lycaon.jsonl --dump $D/rows-sites-lycaon.jsonl \
99
+ --model convaiinnovations/laya-multilingual --out .task/decide/heads/code-rank.safetensors
100
+ scripts/decide/headfile.py show .task/decide/heads/code-rank.safetensors
101
+ ```
102
+
103
+ Labels are the site's own structure on the engine's 0..4 rubric: the target
104
+ 4, another unit in its file 2, a lexical neighbour from another file 1, a
105
+ random candidate 0; synthetic site rows carry explicit levels. The rubric
106
+ text is imported from the engine's serving code so training and serving
107
+ cannot drift. The head file is safetensors with the backbone, label, and
108
+ validation metrics in its header (`scripts/decide/headfile.py`); a GPU host
109
+ trains one in minutes (`--batch-size 64` on an H100), a laptop in hours.
110
+
111
+ The shipped `code-rank` head learns from requests written by various models
112
+ over code units from open-source repositories. It is open weights under
113
+ Apache-2.0, published without its training pairs.
114
+
115
+ ## Results
116
+
117
+ Measured 2026-09-25 on an Apple M1 Pro. Held-out repositories never
118
+ contributed training rows: ast-grep (Rust) and aws-vault (Go). Questions are
119
+ model-written requests describing a unit's purpose without naming it (`claude`
120
+ pairs); doc-derived questions cannot measure doc-bearing candidate text, because
121
+ the request is then the candidate's own comment (lexical MRR 0.94). MRR is the
122
+ mean of 1/rank of the right unit; "+/−" counts questions the engine moved up or
123
+ down; latency is p50 of one engine call through the host's own client with an
124
+ 8 s deadline so no call abstains.
125
+
126
+ **Candidate text matters more than the head.** Adding the signature line
127
+ raised the lexical baseline on ast-grep definitions from MRR 0.134 to 0.193,
128
+ and the leading comment raised it again; both ship regardless of the engine.
129
+ **Zero-shot base models lose to lexical order on every site**; never run the
130
+ base model without a head.
131
+
132
+ **Round three heads** (every site shape, the multi-language corpus, and
133
+ model-written questions in training; multilingual 151k examples, English
134
+ 151k examples, both on an H100). The multilingual head runs through
135
+ `pw-decide`; the English head through the Python runtime, which is two to
136
+ three times faster per candidate than the native engine on this GPU:
137
+
138
+ | head, K | definitions ast-grep n=95 | definitions aws-vault n=54 | repomap ast-grep n=88 | repomap aws-vault n=51 | p50 |
139
+ |---|---|---|---|---|---|
140
+ | multilingual r3, K=16 | 0.096 → 0.120, +16/−7 | 0.135 → 0.163, +9/−2 | 0.100 → 0.125, +11/−4 | 0.121 → 0.133, +5/−2 | 1.1 s |
141
+ | multilingual r3, K=48 | 0.096 → 0.132, +30/−13 | 0.135 → 0.169, +23/−8 | 0.100 → 0.133, +25/−8 | 0.121 → 0.157, +15/−5 | 3.5 s |
142
+ | English r3, K=16 | 0.096 → 0.120, +17/−4 | 0.135 → 0.151, +4/−0 | 0.100 → 0.106, +12/−3 | 0.121 → 0.110, +3/−3 | 0.85 s |
143
+ | English r3, K=48 | 0.108 → 0.139, +27/−17 | 0.132 → 0.153, +13/−9 | 0.100 → 0.137, +26/−6 | 0.121 → 0.136, +14/−9 | 2.5 s |
144
+
145
+ hit@10 moves the same way: multilingual K=48 lifts definitions ast-grep from
146
+ 0.263 to 0.368 and repomap ast-grep from 0.227 to 0.341.
147
+
148
+ **The native engine reproduces the Python runtime exactly**: the same head
149
+ through `pw-decide` and through torch gives identical MRR and identical
150
+ improved/regressed counts (multilingual r3 on MLX, definitions ast-grep:
151
+ 0.096 → 0.132, +30/−13 at K=48 and 0.120, +16/−7 at K=16, the table's rows). Only the
152
+ latency differs: 3.6 s against 1.0 s at K=48, 1.1 s against 0.3 s at K=16,
153
+ because candle's Metal matmul tops out near 2 TFLOPS on this GPU where torch
154
+ reaches five to six.
155
+
156
+ **Project search is hurt by the engine** (file-level target): ast-grep
157
+ 0.777 → 0.623, aws-vault 0.665 → 0.541 with the multilingual r2 head; the
158
+ leg's own score is already strong and narrow, so any blend reorders it. The
159
+ site stays wired and off. **Structure ranking is already solved by lexical
160
+ order** when the comment is in the file head (hit@1 0.95–1.00).
161
+
162
+ **Verdict.** Reranking helps where lexical order is weak (requests that do not
163
+ share words with the unit): +0.02 to +0.04 MRR at K=48, +0.01 to +0.03 at
164
+ K=16, on definitions and repomap pages; it is neutral where lexical order
165
+ already finds the words and harmful on project search. The shipped policy is
166
+ the multilingual checkpoint (half the cost of English, 100+ languages, the
167
+ larger K=48 gains) at K=48 in chunks of 16 with a 2.5 s deadline: on the MLX
168
+ runtime the engine ships with on Apple silicon, definitions on ast-grep at
169
+ K=48 measure 0.096 → 0.132 MRR, +30/−13, at 1.0 s p50 / 1.35 s p95 per call
170
+ with no deadline abstentions and a 1.5 GB engine footprint, where candle on
171
+ Metal needed 3.5 s for the same gain. The unmeasured sites (windows,
172
+ neighbors, call sites, imports, next actions, web pages) stay off until they
173
+ have evaluation pairs; project search stays off because the engine hurts it.
training/turn-load/scripts/decide/rerank/train_rerank.py CHANGED
@@ -1,5 +1,5 @@
1
  #!/usr/bin/env python3
2
- """Fine-tune the Laya decision head on reranking candidates.
3
 
4
  Reads the candidate dumps lycaon/cmd/decide-rerank writes (one row per pair,
5
  every candidate the site offered with its lexical score and target flag) and
@@ -165,7 +165,7 @@ def main():
165
  ap = argparse.ArgumentParser()
166
  ap.add_argument("--dump", action="append", default=[], help="candidate dump JSONL; repeatable")
167
  ap.add_argument("--out", required=True)
168
- ap.add_argument("--model", default=os.environ.get("LAYA_MODEL_ID", "convaiinnovations/laya-multilingual"))
169
  ap.add_argument("--label", default="code-rank")
170
  ap.add_argument("--epochs", type=int, default=6)
171
  ap.add_argument("--batch-size", type=int, default=32)
 
1
  #!/usr/bin/env python3
2
+ """Fine-tune the Painted Wolf Decide head on reranking candidates.
3
 
4
  Reads the candidate dumps lycaon/cmd/decide-rerank writes (one row per pair,
5
  every candidate the site offered with its lexical score and target flag) and
 
165
  ap = argparse.ArgumentParser()
166
  ap.add_argument("--dump", action="append", default=[], help="candidate dump JSONL; repeatable")
167
  ap.add_argument("--out", required=True)
168
+ ap.add_argument("--model", default=os.environ.get("LYCAON_DECIDE_MODEL_ID", "convaiinnovations/laya-multilingual"))
169
  ap.add_argument("--label", default="code-rank")
170
  ap.add_argument("--epochs", type=int, default=6)
171
  ap.add_argument("--batch-size", type=int, default=32)
training/turn-load/scripts/decide/train.py CHANGED
@@ -1,5 +1,5 @@
1
  #!/usr/bin/env python3
2
- """Fine-tune a Laya decision head on turn examples.
3
 
4
  Reads training rows (rows.py, `lycaon-debug decide export`) and trains
5
  the head, scorer, and type embedding of a Laya checkpoint on the same
@@ -61,7 +61,7 @@ import headfile # noqa: E402
61
  import rows as rowfile # noqa: E402
62
  from corpus import Corpus, decide_dir # noqa: E402
63
 
64
- DEFAULT_MODEL = os.environ.get("LAYA_MODEL_ID", "convaiinnovations/laya")
65
 
66
  # Weight on positive options in the multi-label loss: a turn needs a few of its sixty-odd
67
  # loadable tools, and a missed tool costs a round trip where an extra schema costs bytes.
@@ -179,9 +179,28 @@ def tool_weights(rows, mode, tool_truth):
179
  return {name: min(max((total / (n + 1)) ** 0.5, 1.0), 20.0) for name, n in counts.items()}
180
 
181
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
182
  def build_items(examples, corpus, held_units, rng, tok, max_len, head_max_len, families, skill_scored, skill_zeros, turn_rows, pos, tool_truth, observed, tool_negatives=0):
183
- """One training item per (state, question) for the families trained. Tools and guides
184
- are two multi-label items per turn, the kind one choice item, and skills and needs
 
 
185
  score pairs."""
186
  # Choice options in the engine's order: it keys them by name, sorted.
187
  kind_q = {"t": "choice", "ins": corpus.spec["kind"]["instructions"], "crit": dict(sorted(corpus.spec["kind"]["options"].items()))}
@@ -199,27 +218,24 @@ def build_items(examples, corpus, held_units, rng, tok, max_len, head_max_len, f
199
  targets = rowfile.tool_targets(ex, tool_truth)
200
  tools_q = corpus.multi_question("tool", ex["offered"]["loadable"])
201
  if "tools" in families_here and tools_q["options"]:
 
202
  if tools_q.get("independent"):
203
- negatives = [n for n in tools_q["options"] if targets.get(n) == 0]
204
- kept_negatives = set(rng.sample(negatives, min(tool_negatives, len(negatives)))) if tool_negatives else set(negatives)
205
- for name, description in tools_q["options"].items():
206
- if targets.get(name) is None or (targets.get(name) == 0 and name not in kept_negatives):
207
- continue
208
- one = dict(tools_q, options={name: description})
209
- item = multi_item(tok, state, one, {name: targets[name]}, max_len, head_max_len, "tools", host, pos)
210
- if targets[name] == 0:
211
- item["weight"] = [len(negatives) / len(kept_negatives)]
212
- items.append(item)
213
  else:
214
- items.append(multi_item(tok, state, tools_q, {n: targets.get(n) for n in tools_q["options"]}, max_len, head_max_len, "tools", host, pos))
215
  guides = ex["labels"]["guides"]
216
  guides_q = corpus.multi_question("guide", ex["offered"]["guides"])
217
  if "guides" in families_here and guides_q["options"]:
218
  truth = {uid: (None if uid in held_units else guides.get(uid)) for uid in guides_q["options"]}
219
- if any(v is not None for v in truth.values()):
 
 
220
  items.append(multi_item(tok, state, guides_q, truth, max_len, head_max_len, "guides", host))
221
  kind = ex["labels"].get("kind")
222
- if "kind" in families_here and kind in kinds:
223
  ids, markers = build_sequence(tok, state, kind_q, max_len=max_len)
224
  items.append({"ids": ids, "markers": markers, "qtype": QTYPES["choice"], "label": kinds.index(kind), "family": "kind", "host": host})
225
  if "requests" in families_here:
@@ -458,7 +474,8 @@ def main():
458
  ap.add_argument("--rank-levels", choices=("blended", "skills-blended", "judged"), default="blended",
459
  help="blended: a skill read or a tool used after a need trains at 4 whatever the judges said; skills-blended: only a skill read does; judged: the judges' levels alone")
460
  ap.add_argument("--tool-weight", choices=("none", "sqrt-inverse"), default="none", help="per-tool positive weighting")
461
- ap.add_argument("--tool-negatives", type=int, default=0, help="sample this many independent negative tools per training row, with inverse sampling weights; validation keeps all (0 keeps all)")
 
462
  args = ap.parse_args()
463
  if args.tool_negatives < 0:
464
  ap.error("--tool-negatives must be nonnegative")
@@ -473,7 +490,7 @@ def main():
473
 
474
  rng = random.Random(args.seed)
475
  torch.manual_seed(args.seed)
476
- device = os.environ.get("LAYA_DEVICE") or ("cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
477
  corpus = Corpus.load(args.corpus)
478
  held = {uid for uid, u in corpus.units.items() if u["pack_id"] in args.holdout_pack}
479
  if held:
@@ -489,12 +506,9 @@ def main():
489
 
490
  agent = laya.load(args.model, device=device)
491
  tok, model = agent.tok, agent.model
492
- # The engine's option budget: the catalog's head_tokens, clamped to the checkpoint's
493
- # context the way pw-decide clamps it, so training cuts options exactly as serving does.
494
  context = int(agent.cfg.get("max_len", 512))
495
  head_max_len = min(int(corpus.state_spec.get("head_tokens", 512)), max(context - 64, 16))
496
- # The engine encodes every sequence at the checkpoint's full context; training at any
497
- # other length teaches the head on inputs it never sees in service.
498
  max_len = context
499
  print("context %d head budget %d" % (max_len, head_max_len), flush=True)
500
  pos = tool_weights(train, args.tool_weight, args.tool_truth)
@@ -543,8 +557,8 @@ def main():
543
  "corpus": corpus.revision, "holdout": sorted(held), "val_loss": loss, "val_acc": acc, "by_type": by_type,
544
  "train_rows": len(train), "turn_rows": args.turn_rows, "tool_weight": args.tool_weight, "tool_truth": args.tool_truth, "rank_levels": args.rank_levels, "seed": args.seed,
545
  "max_len": max_len, "head_max_len": head_max_len, "pos_weight": POS_WEIGHT, "state_tokens": state_tokens,
546
- "tool_encoding": "independent" if corpus.spec["tools"].get("independent") else "joint", "tool_negatives": args.tool_negatives,
547
- "tool_option_words": corpus.spec["tools"]["option_words"]})
548
  print(" saved %s" % out, flush=True)
549
 
550
 
 
1
  #!/usr/bin/env python3
2
+ """Fine-tune a Painted Wolf Decide head on turn examples.
3
 
4
  Reads training rows (rows.py, `lycaon-debug decide export`) and trains
5
  the head, scorer, and type embedding of a Laya checkpoint on the same
 
61
  import rows as rowfile # noqa: E402
62
  from corpus import Corpus, decide_dir # noqa: E402
63
 
64
+ DEFAULT_MODEL = os.environ.get("LYCAON_DECIDE_MODEL_ID", "convaiinnovations/laya")
65
 
66
  # Weight on positive options in the multi-label loss: a turn needs a few of its sixty-odd
67
  # loadable tools, and a missed tool costs a round trip where an extra schema costs bytes.
 
179
  return {name: min(max((total / (n + 1)) ** 0.5, 1.0), 20.0) for name, n in counts.items()}
180
 
181
 
182
+ def independent_items(tok, state, question, truth, max_len, head_max_len, family, host, pos=None, keep=None, weight=None):
183
+ """One item per option of a multi question, for a head that reads options on their own
184
+ rows. Options without a label are skipped; `keep` names the negative options to keep
185
+ and `weight` the loss weight that restores the sampled negatives' share."""
186
+ items = []
187
+ for name, text in question["options"].items():
188
+ label = truth.get(name)
189
+ if label is None or (not label and keep is not None and name not in keep):
190
+ continue
191
+ one = dict(question, options={name: text})
192
+ item = multi_item(tok, state, one, {name: label}, max_len, head_max_len, family, host, pos)
193
+ if not label and weight is not None:
194
+ item["weight"] = [weight]
195
+ items.append(item)
196
+ return items
197
+
198
+
199
  def build_items(examples, corpus, held_units, rng, tok, max_len, head_max_len, families, skill_scored, skill_zeros, turn_rows, pos, tool_truth, observed, tool_negatives=0):
200
+ """One training item per (state, question) for the families trained. A joint head
201
+ trains tools and guides as two multi-label items per turn and the kind as one choice;
202
+ an independent head trains one item per tool or guide option, with `tool_negatives`
203
+ sampled negative tools per turn (every guide option trains). Skills and needs are
204
  score pairs."""
205
  # Choice options in the engine's order: it keys them by name, sorted.
206
  kind_q = {"t": "choice", "ins": corpus.spec["kind"]["instructions"], "crit": dict(sorted(corpus.spec["kind"]["options"].items()))}
 
218
  targets = rowfile.tool_targets(ex, tool_truth)
219
  tools_q = corpus.multi_question("tool", ex["offered"]["loadable"])
220
  if "tools" in families_here and tools_q["options"]:
221
+ truth = {n: targets.get(n) for n in tools_q["options"]}
222
  if tools_q.get("independent"):
223
+ negatives = [n for n, v in truth.items() if v == 0]
224
+ kept = set(rng.sample(negatives, min(tool_negatives, len(negatives)))) if tool_negatives else set(negatives)
225
+ weight = len(negatives) / len(kept) if kept else None
226
+ items.extend(independent_items(tok, state, tools_q, truth, max_len, head_max_len, "tools", host, pos, kept, weight))
 
 
 
 
 
 
227
  else:
228
+ items.append(multi_item(tok, state, tools_q, truth, max_len, head_max_len, "tools", host, pos))
229
  guides = ex["labels"]["guides"]
230
  guides_q = corpus.multi_question("guide", ex["offered"]["guides"])
231
  if "guides" in families_here and guides_q["options"]:
232
  truth = {uid: (None if uid in held_units else guides.get(uid)) for uid in guides_q["options"]}
233
+ if guides_q.get("independent"):
234
+ items.extend(independent_items(tok, state, guides_q, truth, max_len, head_max_len, "guides", host))
235
+ elif any(v is not None for v in truth.values()):
236
  items.append(multi_item(tok, state, guides_q, truth, max_len, head_max_len, "guides", host))
237
  kind = ex["labels"].get("kind")
238
+ if "kind" in families_here and kind in kinds and not corpus.independent():
239
  ids, markers = build_sequence(tok, state, kind_q, max_len=max_len)
240
  items.append({"ids": ids, "markers": markers, "qtype": QTYPES["choice"], "label": kinds.index(kind), "family": "kind", "host": host})
241
  if "requests" in families_here:
 
474
  ap.add_argument("--rank-levels", choices=("blended", "skills-blended", "judged"), default="blended",
475
  help="blended: a skill read or a tool used after a need trains at 4 whatever the judges said; skills-blended: only a skill read does; judged: the judges' levels alone")
476
  ap.add_argument("--tool-weight", choices=("none", "sqrt-inverse"), default="none", help="per-tool positive weighting")
477
+ ap.add_argument("--tool-negatives", type=int, default=0,
478
+ help="independent heads: sample this many negative tools per training row, weighted to keep their share; validation keeps all (0 keeps all)")
479
  args = ap.parse_args()
480
  if args.tool_negatives < 0:
481
  ap.error("--tool-negatives must be nonnegative")
 
490
 
491
  rng = random.Random(args.seed)
492
  torch.manual_seed(args.seed)
493
+ device = os.environ.get("LYCAON_DECIDE_DEVICE") or ("cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
494
  corpus = Corpus.load(args.corpus)
495
  held = {uid for uid, u in corpus.units.items() if u["pack_id"] in args.holdout_pack}
496
  if held:
 
506
 
507
  agent = laya.load(args.model, device=device)
508
  tok, model = agent.tok, agent.model
509
+ # Match the serving engine's question and context budgets.
 
510
  context = int(agent.cfg.get("max_len", 512))
511
  head_max_len = min(int(corpus.state_spec.get("head_tokens", 512)), max(context - 64, 16))
 
 
512
  max_len = context
513
  print("context %d head budget %d" % (max_len, head_max_len), flush=True)
514
  pos = tool_weights(train, args.tool_weight, args.tool_truth)
 
557
  "corpus": corpus.revision, "holdout": sorted(held), "val_loss": loss, "val_acc": acc, "by_type": by_type,
558
  "train_rows": len(train), "turn_rows": args.turn_rows, "tool_weight": args.tool_weight, "tool_truth": args.tool_truth, "rank_levels": args.rank_levels, "seed": args.seed,
559
  "max_len": max_len, "head_max_len": head_max_len, "pos_weight": POS_WEIGHT, "state_tokens": state_tokens,
560
+ "tool_encoding": "independent" if corpus.independent() else "joint", "tool_negatives": args.tool_negatives,
561
+ "tool_option_words": corpus.spec["tools"]["option_words"], "families": ",".join(families)})
562
  print(" saved %s" % out, flush=True)
563
 
564
 
training/turn-load/train.sh ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ # train.sh GPU RECIPE RUN: train one head into out/RECIPE-RUN, which must not exist (one
3
+ # launch owns a run). Recipes name every argument; data comes from data/ as setup.sh placed it:
4
+ # data/original session rows judged with the first pair (open1's split)
5
+ # data/judged session rows judged by the hosted pair (tools, skills, needs)
6
+ # data/skillreq generated skill requests (train and eval families) judged by the hosted pair
7
+ set -euo pipefail
8
+ SCRIPT_DIR=$(cd -- "$(dirname -- "$0")" && pwd)
9
+ GPU=$1; RECIPE=$2; RUN=$3
10
+ R=${R:-/scratch/bialy}; L=$R/lycaon; PY=$R/venvs/train/bin/python; C=$R/data/corpus.json; OUT=$R/out/$RECIPE-$RUN
11
+ export OMP_NUM_THREADS=${THREADS:-10} MKL_NUM_THREADS=${THREADS:-10} CUDA_VISIBLE_DEVICES=$GPU HF_HOME=$R/hf HF_HUB_OFFLINE=1 \
12
+ LYCAON_DECIDE_MODEL_ID=convaiinnovations/laya-multilingual LYCAON_DECIDE_DEVICE=cuda PYTHONUNBUFFERED=1
13
+ mkdir -p $R/out
14
+ mkdir $OUT 2>/dev/null || { echo "$OUT exists; choose a new run name" >&2; exit 1; }
15
+ cd $L
16
+ TURN="--holdout-pack painted-wolf/browser --families tools,guides,kind --seed 11 --batch-size 32"
17
+ RANK="--families skills,requests --skill-scored 4 --skill-zeros 3 --seed 11 --batch-size 32"
18
+ case $RECIPE in
19
+ # B7 trains the B5 recipe with the guide units beside the tools, every option on its
20
+ # own row; B7G trains the guide units alone, as the fallback head for a second slot.
21
+ B7|B7G)
22
+ C=$R/data/independent-corpus.json
23
+ FAMILIES=tools,guides; [[ $RECIPE = B7G ]] && FAMILIES=guides
24
+ $PY -c 'import json,sys; c=json.load(open(sys.argv[1])); assert c["questions"]["tools"].get("independent"), "independent corpus required"' "$C"
25
+ exec $PY scripts/decide/train.py --corpus "$C" --train "$R/data/judged/train.jsonl" --val "$R/data/judged/val.jsonl" \
26
+ --families "$FAMILIES" --tool-truth consensus --tool-weight none --tool-negatives 24 --lr 5e-4 --pos-weight 6 --seed 11 \
27
+ --batch-size 64 --epochs 45 --patience 8 --label "open1-turn-load-$RECIPE-$RUN-independent" --out "$OUT/turn-load.safetensors" ;;
28
+ B3|B4|B5|B6)
29
+ C=$R/data/independent-corpus.json
30
+ TRUTH=consensus; [[ $RECIPE = B4 ]] && TRUTH=called
31
+ NEGATIVES=8; LR=5e-4
32
+ [[ $RECIPE = B5 || $RECIPE = B6 ]] && NEGATIVES=24
33
+ [[ $RECIPE = B6 ]] && LR=1e-4
34
+ $PY -c 'import json,sys; c=json.load(open(sys.argv[1])); assert c["questions"]["tools"].get("independent"), "independent corpus required"' "$C"
35
+ exec $PY scripts/decide/train.py --corpus "$C" --train "$R/data/judged/train.jsonl" --val "$R/data/judged/val.jsonl" \
36
+ --families tools --tool-truth "$TRUTH" --tool-weight none --tool-negatives "$NEGATIVES" --lr "$LR" --pos-weight 6 --seed 11 \
37
+ --batch-size 64 --epochs 45 --patience 8 --label "open1-turn-load-$RECIPE-$RUN-independent" --out "$OUT/turn-load.safetensors" ;;
38
+ # Turn-load. A2 is rev4a's recipe on the open sessions; B2 the same on judged labels.
39
+ A2) exec $PY scripts/decide/train.py --corpus $C --train $R/data/original/train.jsonl --val $R/data/original/val.jsonl $TURN \
40
+ --tool-weight none --tool-truth called --label turn-load-a2 --out $OUT/turn-load.safetensors ;;
41
+ B2) exec $PY scripts/decide/train.py --corpus $C --train $R/data/judged/train.jsonl --val $R/data/judged/val.jsonl $TURN \
42
+ --tool-weight none --tool-truth consensus --label turn-load-b2 --out $OUT/turn-load.safetensors ;;
43
+ # Unit-rank. E counts a session's skill reads over the judges; E1 and E2 add the generated
44
+ # skill requests, all of them or half the families.
45
+ E|E0) exec $PY scripts/decide/train.py --corpus $C --train $R/data/judged/train.jsonl --val $R/data/selection.jsonl $RANK \
46
+ --rank-levels skills-blended --label "open1-unit-rank-$(echo "$RECIPE" | tr A-Z a-z)-$RUN" --out $OUT/unit-rank.safetensors ;;
47
+ E6) exec $PY scripts/decide/train.py --corpus $C --train $R/derived/hard-skill-train.jsonl --val $R/data/selection.jsonl \
48
+ --families skills,requests --skill-scored 8 --skill-zeros 12 --seed 11 --batch-size 32 \
49
+ --rank-levels skills-blended --label "open1-unit-rank-e6-$RUN" --out $OUT/unit-rank.safetensors ;;
50
+ E1|E2|E3|E4|E5)
51
+ FRACTION=0.5; [[ $RECIPE = E1 || $RECIPE = E3 ]] && FRACTION=1.0
52
+ LEVELS=skills-blended
53
+ if [[ $RECIPE = E3 || $RECIPE = E4 || $RECIPE = E5 ]]; then
54
+ RANK="--families skills,requests --skill-scored 8 --skill-zeros 12 --seed 11 --batch-size 32"
55
+ fi
56
+ [[ $RECIPE = E5 ]] && LEVELS=judged
57
+ $PY "$SCRIPT_DIR/mix.py" $OUT/train.jsonl $FRACTION 7 $R/data/judged/train.jsonl $R/data/skillreq/train.judged.jsonl
58
+ exec $PY scripts/decide/train.py --corpus $C --train $OUT/train.jsonl --val $R/data/selection.jsonl $RANK \
59
+ --rank-levels "$LEVELS" --label "open1-unit-rank-$(echo "$RECIPE" | tr A-Z a-z)-$RUN" --out $OUT/unit-rank.safetensors ;;
60
+ *) echo "unknown recipe $RECIPE" >&2; exit 2 ;;
61
+ esac
training/turn-load/turn_probe.py ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Score every option of a turn head with the trainer's own forward, tools and guides alike.
2
+
3
+ Writes one row per complete example: the row's key, offered candidates and labels, and
4
+ the head's probability for every tool and guide option, encoded exactly as the engine
5
+ encodes them (one option per row for an independent head, the roster for a joint one).
6
+ The output feeds `guide_audit.py` and `turn_score.py` without an engine build.
7
+
8
+ Usage: turn_probe.py --trainer DIR --corpus FILE --rows FILE --head FILE --out FILE [--limit N]
9
+ """
10
+ import argparse
11
+ import hashlib
12
+ import json
13
+ import os
14
+ import sys
15
+ from pathlib import Path
16
+
17
+
18
+ def option_probs(agent, model, corpus, ex, device, kind, qid, family, torch, multi_item, precompute, forward_head, state_text):
19
+ """Probabilities per option of one multi question, encoded as the engine would."""
20
+ context = int(agent.cfg.get("max_len", 512))
21
+ head_max_len = min(int(corpus.state_spec.get("head_tokens", 512)), context - 64)
22
+ question = corpus.multi_question(kind, ex["offered"]["loadable" if kind == "tool" else "guides"])
23
+ if not question["options"]:
24
+ return {}
25
+ state = state_text(ex["state"])
26
+ if question.get("independent"):
27
+ items, names = [], []
28
+ for name, text in question["options"].items():
29
+ one = dict(question, options={name: text})
30
+ items.append(multi_item(agent.tok, state, one, {name: 0}, context, head_max_len, family))
31
+ names.append(name)
32
+ else:
33
+ item = multi_item(agent.tok, state, question, {k: 0 for k in question["options"]}, context, head_max_len, family)
34
+ items, names = [item], list(question["options"].keys())[: len(item["markers"])]
35
+ features = precompute(agent, items, device, batch_size=64)
36
+ out = {}
37
+ with torch.no_grad():
38
+ for f, item in zip(features, items):
39
+ logits = forward_head(model, f["h"][None].to(device), f["att"][None].to(device), f["marker_pos"][None].to(device),
40
+ f["marker_mask"][None].to(device), torch.tensor([item["qtype"]], device=device))[0].cpu()
41
+ probs = torch.sigmoid(logits[: len(item["markers"])]).tolist()
42
+ if question.get("independent"):
43
+ out[names[len(out)]] = probs[0]
44
+ else:
45
+ out.update(zip(names, probs))
46
+ return out
47
+
48
+
49
+ def main():
50
+ parser = argparse.ArgumentParser(description=__doc__)
51
+ for name in ("trainer", "corpus", "rows", "head", "out"):
52
+ parser.add_argument("--" + name, required=True)
53
+ parser.add_argument("--model", default="convaiinnovations/laya-multilingual")
54
+ parser.add_argument("--limit", type=int, default=0, help="complete rows to score; 0 scores every complete row")
55
+ args = parser.parse_args()
56
+ sys.path.insert(0, str(Path(args.trainer) / "scripts/decide"))
57
+ import laya
58
+ import torch
59
+ from corpus import Corpus
60
+ from headfile import read_metadata
61
+ from rows import load
62
+ from safetensors.torch import load_file
63
+ from train import forward_head, multi_item, precompute, state_text
64
+
65
+ torch.set_num_threads(8)
66
+ device = os.environ.get("LYCAON_DECIDE_DEVICE", "cpu")
67
+ corpus = Corpus.load(args.corpus)
68
+ agent = laya.load(args.model, device=device)
69
+ meta = read_metadata(args.head)[0]
70
+ context = int(agent.cfg.get("max_len", 512))
71
+ budget = min(int(corpus.state_spec.get("head_tokens", 512)), context - 64)
72
+ encoding = "independent" if corpus.independent() else "joint"
73
+ if (meta.get("max_len"), meta.get("head_max_len"), meta.get("tool_encoding", "joint")) != (str(context), str(budget), encoding):
74
+ raise ValueError("head encoding metadata does not match inference: %s" % {k: meta.get(k) for k in ("max_len", "head_max_len", "tool_encoding")})
75
+ tensors = load_file(args.head, device=device)
76
+ for name in ("head", "scorer", "type_emb"):
77
+ getattr(agent.model, name).load_state_dict({k[len(name) + 1:]: v for k, v in tensors.items() if k.startswith(name + ".")})
78
+ agent.model.eval()
79
+ examples = [r for r in load(args.rows) if not r["partial"]]
80
+ if args.limit:
81
+ examples = examples[: args.limit]
82
+ families = set((meta.get("families") or "tools,guides").split(","))
83
+ signature = {name: hashlib.sha256(Path(getattr(args, name)).read_bytes()).hexdigest() for name in ("head", "corpus", "rows")}
84
+ helpers = (torch, multi_item, precompute, forward_head, state_text)
85
+ with Path(args.out).open("x") as out:
86
+ for n, row in enumerate(examples, 1):
87
+ answers = {}
88
+ if "tools" in families:
89
+ answers["tools"] = {"probabilities": option_probs(agent, agent.model, corpus, row, device, "tool", "tools", "tools", *helpers)}
90
+ if "guides" in families:
91
+ answers["guides"] = {"probabilities": option_probs(agent, agent.model, corpus, row, device, "guide", "guides", "guides", *helpers)}
92
+ json.dump({"key": "%s:%s" % (row["session"], row["receipt"]), "host": row["host"], "offered": row["offered"],
93
+ "labels": {"tools": row["labels"].get("tools", []), "guides": row["labels"].get("guides", {})},
94
+ "answers": answers}, out)
95
+ out.write("\n")
96
+ if n % 100 == 0:
97
+ out.flush()
98
+ sys.stderr.write("scored %d/%d\n" % (n, len(examples)))
99
+ Path(args.out + ".meta.json").write_text(json.dumps({"sha256": signature, "encoding": meta, "device": device, "rows": len(examples)}, indent=2))
100
+
101
+
102
+ if __name__ == "__main__":
103
+ main()
val.jsonl CHANGED
The diff for this file is too large to render. See raw diff