Hugging Face's logo

modrill
/
math-think-o7b-20260908

Text Generation
Transformers
Safetensors
olmo3
math
sft
lora
think
task-vector
iclr2027
conversational
Model card Files
xet
 Community
modrill commited on
Commit
7fdfe05
·
verified ·
1 Parent(s): cd83605

freeze: Think 2ep merged endpoint + LoRA adapter for modrill/think-src-o7b-20260908

Browse files
Files changed (12) hide show
  1. COMMIT.json +100 -0
  2. MERGE_RECEIPT.json +19 -0
  3. README.md +55 -0
  4. TOKEN_BOUNDARY_LEDGER.json +0 -0
  5. adapter/adapter_config.json +57 -0
  6. adapter/adapter_model.safetensors +3 -0
  7. chat_template.jinja +4 -0
  8. config.json +75 -0
  9. generation_config.json +4 -0
  10. model.safetensors +3 -0
  11. tokenizer.json +0 -0
  12. tokenizer_config.json +13 -0
COMMIT.json ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema": "ddc-atomic-checkpoint-commit/1",
3
+ "runner_id": "THINK_B_2EP_SIX_ARMS_RUNNER_V3",
4
+ "completed_optimizer_updates": 987,
5
+ "cumulative_realized_active_tokens": 64632872,
6
+ "bindings": {
7
+ "config_sha256": "5c5965a584fe72804b4752efb2e27b010079eb50e0357eb288923d913e458fa4",
8
+ "runner_code_sha256": "fd6d1c6403810acc4460e2ce0fb5f4418881387b11f4bd21f7cd21067faa3a0a",
9
+ "config_schema_sha256": "d6d13eeee42bf04cf34fc68b2fa7a61b316c4543b77d7d779815b8eddf10c099",
10
+ "python_binary_sha256": "5700a6a7804feb44da6c963ebd1323982478b7ca57478e4e81981c47818453bd",
11
+ "environment_lock_sha256": "627d4fe088b8f64841b0520ae5bd9dc5c81e9dcd5969b4bd45c9a9910b2f3c01",
12
+ "base_composite_sha256": "526963d72561cb6244f60f751312673d2671111d214636ae571de0d62c762760",
13
+ "base_lock_sha256": "6791648e4030008388f2e9241b23771cb07eb0416fbd7bc3c6777c7c36bda3be",
14
+ "frozen_rows_sha256": "8221ff30de8fa345441e6002c111c948a3fdd20413e5c09e3a8f76cf26b12ef1",
15
+ "ordered_ids_sha256": "dde87a60437027a3e19566382b4ac25381f2888e45507bef0ddc3d6d284c412b",
16
+ "update_plan_sha256": "24e1f6fecc9e9e5b5c371c7041cbe0bf79dec48741caf8a60d50310984e59ef1",
17
+ "data_receipt_sha256": "e66346fb413557929ab53648bc656fb3102557cf3460abb49c7b1c6e7281420d",
18
+ "decontamination_receipt_sha256": "66c360871b06208ed20e473e57ad0f09c79e4dac1b3124a9461a5b729b487153",
19
+ "render_receipt_sha256": "e66346fb413557929ab53648bc656fb3102557cf3460abb49c7b1c6e7281420d",
20
+ "token_bake_sha256": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b",
21
+ "continuation_sha256": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b"
22
+ },
23
+ "trainer_state": {
24
+ "path": "trainer_state.pt",
25
+ "bytes": 1919557677,
26
+ "sha256": "1ecc81d3f7ee1f1acffd3a9cba3ccf42de56b7bc476571a1b928dee3d2d7058b"
27
+ },
28
+ "token_boundary_ledger": {
29
+ "path": "TOKEN_BOUNDARY_LEDGER.json",
30
+ "bytes": 970601,
31
+ "sha256": "bbf8fc19c70dc5bab6b95bfd8799a022599968157ab72f4e3dc0c1f196e74a4b"
32
+ },
33
+ "adapter_files": [
34
+ {
35
+ "path": "adapter/README.md",
36
+ "bytes": 5282,
37
+ "sha256": "45145483fe203512c1f26bd8675f16875cc326c7cfe5e3020cdf0f50f297acf2"
38
+ },
39
+ {
40
+ "path": "adapter/adapter_config.json",
41
+ "bytes": 1276,
42
+ "sha256": "b0b7fe4369c25edd35ad553eb42a78160c4317ba719fbc1ece7cf824dae037bb"
43
+ },
44
+ {
45
+ "path": "adapter/adapter_model.safetensors",
46
+ "bytes": 639724904,
47
+ "sha256": "9da5899df6a1d6d7cfb9e412085eed9a85693bd0fd848645b176e34deed0d8e2"
48
+ }
49
+ ],
50
+ "serialization_canary": {
51
+ "save_load_format": "safetensors",
52
+ "state_key_count": 450,
53
+ "token_state": {
54
+ "schema": "think-b-2ep-token-rows/1",
55
+ "tensors": [
56
+ {
57
+ "key": "base_model.model.lm_head.token_adapter.trainable_tokens_delta",
58
+ "sha256": "037ef35dc1d9a3f53df04a14fc8f4c259ae071ac569f124410b850e091ada551",
59
+ "shape": [
60
+ 1,
61
+ 4096
62
+ ],
63
+ "dtype": "torch.float32"
64
+ },
65
+ {
66
+ "key": "base_model.model.model.embed_tokens.token_adapter.trainable_tokens_delta",
67
+ "sha256": "1f1a38b5c91a31be7cc7cd0f4dded40c8cb9781f2ceaefda73c2eb5e2d52e899",
68
+ "shape": [
69
+ 1,
70
+ 4096
71
+ ],
72
+ "dtype": "torch.float32"
73
+ }
74
+ ]
75
+ }
76
+ },
77
+ "token_state": {
78
+ "schema": "think-b-2ep-token-rows/1",
79
+ "tensors": [
80
+ {
81
+ "key": "base_model.model.lm_head.token_adapter.trainable_tokens_delta",
82
+ "sha256": "037ef35dc1d9a3f53df04a14fc8f4c259ae071ac569f124410b850e091ada551",
83
+ "shape": [
84
+ 1,
85
+ 4096
86
+ ],
87
+ "dtype": "torch.float32"
88
+ },
89
+ {
90
+ "key": "base_model.model.model.embed_tokens.token_adapter.trainable_tokens_delta",
91
+ "sha256": "1f1a38b5c91a31be7cc7cd0f4dded40c8cb9781f2ceaefda73c2eb5e2d52e899",
92
+ "shape": [
93
+ 1,
94
+ 4096
95
+ ],
96
+ "dtype": "torch.float32"
97
+ }
98
+ ]
99
+ }
100
+ }
MERGE_RECEIPT.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema": "q4b-think-b-2ep-merge-receipt/1",
3
+ "arm": "O7B-THINK-A3B",
4
+ "checkpoint": "/workspace/math_think_a3b2507_20260906/train_v3_eot_20260908/runs/O7B-THINK-A3B/train/checkpoint-update-00000987-tokens-000064632872",
5
+ "device": "cuda:0",
6
+ "dtype": "bfloat16",
7
+ "adapter_config_sha256": "b0b7fe4369c25edd35ad553eb42a78160c4317ba719fbc1ece7cf824dae037bb",
8
+ "adapter_model_sha256": "9da5899df6a1d6d7cfb9e412085eed9a85693bd0fd848645b176e34deed0d8e2",
9
+ "trainable_token_indices": {
10
+ "embed_tokens": [
11
+ 100257
12
+ ],
13
+ "lm_head": [
14
+ 100257
15
+ ]
16
+ },
17
+ "untied": true,
18
+ "tokenizer_source": "/workspace/OLMO3-MATH-DUAL-v1/models/Olmo-3-1025-7B-996971efdc50"
19
+ }
README.md ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: allenai/Olmo-3-1025-7B
4
+ library_name: transformers
5
+ tags:
6
+ - math
7
+ - sft
8
+ - lora
9
+ - think
10
+ - task-vector
11
+ - iclr2027
12
+ ---
13
+
14
+ # think-src-o7b-20260908
15
+
16
+ Public freeze of Math Think **2ep endpoint** for ICLR 2027 task-vector work.
17
+ Not a chatbot. **Endpoint is the score.** Do not promote 1ep / 1.5ep milestones.
18
+
19
+ `run_id=math_six_arms_train_v3_eot_20260908`. Sibling of `modrill/nothink-src-*-20260908`; does not overwrite those repos.
20
+
21
+ ## Score (Exact-240)
22
+
23
+ AIME24+25 × seeds 42–45, EvalScope reviews, n=240. Authority field `evalscope_reviews.correct`.
24
+
25
+ | Model | Official /240 | cap |
26
+ |---|---:|---:|
27
+ | This 2ep endpoint | **51** | 124 |
28
+ | Same-run Think Base | 34 | |
29
+
30
+ Think mode; stop ids [100257, 100265]; O7B eval uses OLMO3-VLLM-PATCH-v1 / vLLM 0.27.1. Same-run Think Base is think_vllm027 34/240 (first think Base attempt failed).
31
+
32
+ θ_0 is `allenai/Olmo-3-1025-7B` rev `996971efdc504b81f0a6caf73a6c92f976254b9c`.
33
+
34
+ ## Recipe
35
+
36
+ OpenR1 11750 rows × 2ep (unique problems 5875), LoRA r64/α128, lr 1e-4, TPU 65536, seed 42, EOT (Qwen tail `151643` / O7B tail `100257`), B-rows both sides [100257].
37
+
38
+ ## Identity
39
+
40
+ | Field | Value |
41
+ |---|---|
42
+ | Arm | `O7B-THINK-A3B` |
43
+ | Updates / tokens | 987 / 64,632,872 |
44
+ | Merged `model.safetensors` sha256 | `9f75574aed1be033f277a0384b97f4edef3f8ac85401c6781e59c1d46737d526` |
45
+ | Endpoint adapter sha256 | `9da5899df6a1d6d7cfb9e412085eed9a85693bd0fd848645b176e34deed0d8e2` |
46
+
47
+ Root of merged weights is this repo. Endpoint LoRA is in `adapter/`. `MERGE_RECEIPT.json` is the merge audit. `trainer_state` is not uploaded.
48
+
49
+ ## Load
50
+
51
+ ```python
52
+ from transformers import AutoModelForCausalLM, AutoTokenizer
53
+ m = AutoModelForCausalLM.from_pretrained("modrill/think-src-o7b-20260908", torch_dtype="bfloat16", device_map="auto")
54
+ tok = AutoTokenizer.from_pretrained("modrill/think-src-o7b-20260908")
55
+ ```
TOKEN_BOUNDARY_LEDGER.json ADDED
The diff for this file is too large to render. See raw diff
 
adapter/adapter_config.json ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "/workspace/OLMO3-MATH-DUAL-v1/models/Olmo-3-1025-7B-996971efdc50",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 128,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.0,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "monteclora_config": null,
27
+ "peft_type": "LORA",
28
+ "peft_version": "0.20.0",
29
+ "qalora_group_size": 16,
30
+ "r": 64,
31
+ "rank_pattern": {},
32
+ "revision": null,
33
+ "target_modules": [
34
+ "gate_proj",
35
+ "q_proj",
36
+ "v_proj",
37
+ "down_proj",
38
+ "up_proj",
39
+ "k_proj",
40
+ "o_proj"
41
+ ],
42
+ "target_parameters": null,
43
+ "task_type": "CAUSAL_LM",
44
+ "trainable_token_indices": {
45
+ "embed_tokens": [
46
+ 100257
47
+ ],
48
+ "lm_head": [
49
+ 100257
50
+ ]
51
+ },
52
+ "use_bdlora": null,
53
+ "use_dora": false,
54
+ "use_qalora": false,
55
+ "use_rslora": false,
56
+ "velora_config": null
57
+ }
adapter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9da5899df6a1d6d7cfb9e412085eed9a85693bd0fd848645b176e34deed0d8e2
3
+ size 639724904
chat_template.jinja ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {% for message in messages %}{{'<|im_start|>' + message['role'] + '
2
+ ' + message['content'] + '<|im_end|>' + '
3
+ '}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
4
+ ' }}{% endif %}
config.json ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Olmo3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "bfloat16",
9
+ "eos_token_id": 100257,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 4096,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 11008,
14
+ "layer_types": [
15
+ "sliding_attention",
16
+ "sliding_attention",
17
+ "sliding_attention",
18
+ "full_attention",
19
+ "sliding_attention",
20
+ "sliding_attention",
21
+ "sliding_attention",
22
+ "full_attention",
23
+ "sliding_attention",
24
+ "sliding_attention",
25
+ "sliding_attention",
26
+ "full_attention",
27
+ "sliding_attention",
28
+ "sliding_attention",
29
+ "sliding_attention",
30
+ "full_attention",
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "sliding_attention",
34
+ "full_attention",
35
+ "sliding_attention",
36
+ "sliding_attention",
37
+ "sliding_attention",
38
+ "full_attention",
39
+ "sliding_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "full_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "sliding_attention",
46
+ "full_attention"
47
+ ],
48
+ "max_position_embeddings": 65536,
49
+ "model_type": "olmo3",
50
+ "num_attention_heads": 32,
51
+ "num_hidden_layers": 32,
52
+ "num_key_value_heads": 32,
53
+ "pad_token_id": 100277,
54
+ "rms_norm_eps": 1e-06,
55
+ "rope_parameters": {
56
+ "full_attention": {
57
+ "attention_factor": 1.2079441541679836,
58
+ "beta_fast": 32,
59
+ "beta_slow": 1,
60
+ "factor": 8.0,
61
+ "original_max_position_embeddings": 8192,
62
+ "rope_theta": 500000,
63
+ "rope_type": "yarn"
64
+ },
65
+ "sliding_attention": {
66
+ "rope_theta": 500000.0,
67
+ "rope_type": "default"
68
+ }
69
+ },
70
+ "sliding_window": 4096,
71
+ "tie_word_embeddings": false,
72
+ "transformers_version": "5.15.0",
73
+ "use_cache": true,
74
+ "vocab_size": 100278
75
+ }
generation_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "transformers_version": "5.15.0"
4
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f75574aed1be033f277a0384b97f4edef3f8ac85401c6781e59c1d46737d526
3
+ size 14596063960
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|endoftext|>",
7
+ "is_local": true,
8
+ "local_files_only": true,
9
+ "model_max_length": 65536,
10
+ "pad_token": "<|pad|>",
11
+ "tokenizer_class": "TokenizersBackend",
12
+ "unk_token": "<|endoftext|>"
13
+ }