Hugging Face's logo

Datasets:
thoughtworks
/
arithmetic-sorl-data

Tasks:
Text Generation
Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
1M - 10M
ArXiv:
Tags:
arithmetic
interpretability
sorl
mechanistic-interpretability
addition
subtraction
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
amirali1985 commited on
Commit
29d1f0a
·
verified ·
1 Parent(s): cd802fb

Initial data catalog: 10 entries

Browse files
Files changed (1) hide show
  1. data_catalog.json +102 -0
data_catalog.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "path": "add_6digit/eval_stratified.parquet",
4
+ "description": "Stratified eval: legacy addition-only, superseded",
5
+ "version": "v1",
6
+ "n_examples": null,
7
+ "n_splits": null,
8
+ "seed": null,
9
+ "status": "DEPRECATED",
10
+ "notes": ""
11
+ },
12
+ {
13
+ "path": "add_6digit/train.parquet",
14
+ "description": "Training data: 6-digit addition only",
15
+ "version": "v1",
16
+ "n_examples": null,
17
+ "n_splits": null,
18
+ "seed": null,
19
+ "status": "ACTIVE",
20
+ "notes": ""
21
+ },
22
+ {
23
+ "path": "add_6digit/val.parquet",
24
+ "description": "Validation data: 6-digit addition only",
25
+ "version": "v1",
26
+ "n_examples": null,
27
+ "n_splits": null,
28
+ "seed": null,
29
+ "status": "ACTIVE",
30
+ "notes": ""
31
+ },
32
+ {
33
+ "path": "add_handcrafted/test.parquet",
34
+ "description": "Handcrafted addition test cases",
35
+ "version": "v1",
36
+ "n_examples": null,
37
+ "n_splits": null,
38
+ "seed": null,
39
+ "status": "ACTIVE",
40
+ "notes": ""
41
+ },
42
+ {
43
+ "path": "add_sub_6digit/eval_stratified.parquet",
44
+ "description": "Stratified eval: legacy format, superseded by JSON eval sets",
45
+ "version": "v1",
46
+ "n_examples": null,
47
+ "n_splits": null,
48
+ "seed": null,
49
+ "status": "DEPRECATED",
50
+ "notes": "Use eval_sets/eval_add_sub_6d_N100_seed42.json instead."
51
+ },
52
+ {
53
+ "path": "add_sub_6digit/train.parquet",
54
+ "description": "Training data: 6-digit add/sub with Quirke enrichment",
55
+ "version": "v1",
56
+ "n_examples": null,
57
+ "n_splits": null,
58
+ "seed": null,
59
+ "status": "ACTIVE",
60
+ "notes": ""
61
+ },
62
+ {
63
+ "path": "add_sub_6digit/val.parquet",
64
+ "description": "Validation data: 6-digit add/sub",
65
+ "version": "v1",
66
+ "n_examples": null,
67
+ "n_splits": null,
68
+ "seed": null,
69
+ "status": "ACTIVE",
70
+ "notes": ""
71
+ },
72
+ {
73
+ "path": "eval_sets/eval_add_sub_6d_N100_seed42.json",
74
+ "description": "Canonical final eval set: 100 examples per Quirke split, seed=42",
75
+ "version": "v1",
76
+ "n_examples": 2200,
77
+ "n_splits": 22,
78
+ "seed": 42,
79
+ "status": "ACTIVE",
80
+ "notes": "Used for all final eval comparisons. Downloaded by get_eval_set()."
81
+ },
82
+ {
83
+ "path": "eval_sets/eval_add_sub_6d_N25_seed42.json",
84
+ "description": "Epoch eval set: 25 examples per split for mid-training checks",
85
+ "version": "v1",
86
+ "n_examples": 550,
87
+ "n_splits": 22,
88
+ "seed": 42,
89
+ "status": "ACTIVE",
90
+ "notes": "Used during training for per-epoch accuracy. Downloaded by get_epoch_eval_set()."
91
+ },
92
+ {
93
+ "path": "sub_handcrafted/test.parquet",
94
+ "description": "Handcrafted subtraction test cases",
95
+ "version": "v1",
96
+ "n_examples": null,
97
+ "n_splits": null,
98
+ "seed": null,
99
+ "status": "ACTIVE",
100
+ "notes": ""
101
+ }
102
+ ]