Hugging Face's logo

Siyuc
/
loopTF-workspace

Safetensors
Model card Files
xet
 Community
Siyuc commited on
Commit
bf78c1b
·
verified ·
1 Parent(s): a6ac675

Add Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1 step 30520 weights

Browse files
Files changed (8) hide show
  1. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/README.md +19 -0
  2. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/config.json +342 -0
  3. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/experiment.yaml +115 -0
  4. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/manifest.json +37 -0
  5. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/model.safetensors +3 -0
  6. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/tokenizer/special_tokens_map.json +8 -0
  7. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/tokenizer/tokenizer.json +0 -0
  8. Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/tokenizer/tokenizer_config.json +45 -0
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/README.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1 — step 30520
2
+
3
+ Final loop_budget weights; 50,436,160 parameters; 1,000,079,360 training tokens.
4
+
5
+ Model tensors were consolidated on CPU, checked for finite values and exact safetensors equality, and strictly loaded before and after serialization. Optimizer/RNG state and training data are excluded. These custom architectures use the project GridTask loader, not Transformers AutoModel or the Megatron adapter.
6
+
7
+ Loading revision: `8f346b4b28cb8289ccbb50f3435fbfcaf08d8202` with loop_budget support.
8
+
9
+ ```python
10
+ from pathlib import Path
11
+ from safetensors.torch import load_file
12
+ from loop_transformer_backends.loop_budget import GridTask, load_config
13
+
14
+ folder = Path('/path/to/downloaded/step_folder')
15
+ config = load_config(folder / 'experiment.yaml')
16
+ model = GridTask(config).build_model()
17
+ model.load_state_dict(load_file(str(folder / 'model.safetensors')), strict=True, assign=True)
18
+ model.eval()
19
+ ```
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/config.json ADDED
@@ -0,0 +1,342 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architecture": "loop_budget",
3
+ "budget_config": {
4
+ "backprop_steps": 3,
5
+ "coda": 5,
6
+ "core": 10,
7
+ "entropy_beta": 0.1,
8
+ "feedback_loss_weight": 1.0,
9
+ "huginn_intermediate_size": null,
10
+ "inner_loops": 3,
11
+ "kind": "huginn",
12
+ "lookahead_auxiliary_loss_passes": "final",
13
+ "lookahead_auxiliary_loss_weighting": "sum",
14
+ "lookahead_draft_loss_weight": 0.0,
15
+ "lookahead_future_loss_weight": 0.0,
16
+ "lookahead_refiner_intermediate_size": null,
17
+ "lookahead_refiner_layers": 4,
18
+ "normalize_outer_loss": true,
19
+ "ouro_objective": "stage1",
20
+ "outer_passes": 3,
21
+ "parameter_tolerance": 0.01,
22
+ "prelude": 5,
23
+ "target_parameters": 50000000
24
+ },
25
+ "checkpoint_format": "safetensors_loop_budget_state_dict",
26
+ "checkpoint_iteration": 30520,
27
+ "execution_architecture": "loop_budget",
28
+ "experiment_config": "experiment.yaml",
29
+ "final_checkpoint": true,
30
+ "final_validation": {
31
+ "val_full_depth_loss": 3.020910395309329
32
+ },
33
+ "iteration": 30520,
34
+ "loading_code_revision": "8f346b4b28cb8289ccbb50f3435fbfcaf08d8202",
35
+ "loading_note": "Use GridTask/load_config as documented. This is not a Megatron or Transformers AutoModel export.",
36
+ "model_config": {
37
+ "attention_dropout": 0.0,
38
+ "attn_implementation": "sdpa",
39
+ "backbone_profile": "paper2604_vanilla",
40
+ "gradient_checkpointing": true,
41
+ "hf_config_overrides": {},
42
+ "hidden_size": 320,
43
+ "initializer_range": 0.02,
44
+ "intermediate_size": 1664,
45
+ "max_position_embeddings": 2048,
46
+ "model_type": "full_bandwidth",
47
+ "num_attention_heads": 5,
48
+ "num_hidden_layers": 20,
49
+ "num_key_value_heads": 5,
50
+ "padded_vocab_size": 32064,
51
+ "paper_allow_ffn_override": true,
52
+ "paper_full_attention_interval": 6,
53
+ "paper_sliding_window": 2048,
54
+ "pretrained_model_name_or_path": null,
55
+ "revision": null,
56
+ "rms_norm_eps": 1e-06,
57
+ "rope_theta": 10000.0,
58
+ "tie_word_embeddings": false,
59
+ "trust_remote_code": false,
60
+ "vocab_size": 32001
61
+ },
62
+ "model_kind": "huginn",
63
+ "parameter_counts": {
64
+ "total_parameters": 50436160,
65
+ "total_trainable": 50436160
66
+ },
67
+ "provenance": {
68
+ "queue_run_id": "20260923-k3-lr-bs-direct-r2",
69
+ "recovery_revision": null,
70
+ "source_snapshot_hashes": {
71
+ "snapshot/src/full_bandwidth_loop_tf/__init__.py": "7d7213a2ad8c47a51bd619e58185bd6a77bbb54dd3d42c098031e6bb980d7994",
72
+ "snapshot/src/full_bandwidth_loop_tf/adapter.py": "f7979fb2c58d185aea94af08c042798c1e18c690331363bde05d13f6e595533b",
73
+ "snapshot/src/full_bandwidth_loop_tf/adapter_policy.py": "7944aedde11852f2821e77d325c6963f3537ac18eb7d7a6e9deee38867ef7193",
74
+ "snapshot/src/full_bandwidth_loop_tf/hf/__init__.py": "f9f4f6d9cfaa80a0443983c70e46cec8b9361ea69a75667717eb8df1c201cc74",
75
+ "snapshot/src/full_bandwidth_loop_tf/hf/backbone.py": "0eafcedbf30552ea910628b5e845ecb73d9dc2daca99d73e38748d873e3b42dc",
76
+ "snapshot/src/full_bandwidth_loop_tf/hf/config.py": "f82ca8335df23b35b22ba63d0bc42437c97e3cb0d960943b6608e3c1cf4f6827",
77
+ "snapshot/src/full_bandwidth_loop_tf/hf/model.py": "21219220c8c4313fb8a6290bf5d492631f135dba679af407151b3f6ca924e683",
78
+ "snapshot/src/full_bandwidth_loop_tf/hf/plugin.py": "2b15947ce20c8be1676551ddcce6be9bf65d0c3f200064b06fe0ae7460a7f4b4",
79
+ "snapshot/src/full_bandwidth_loop_tf/hf/task.py": "400c0869028798ed1badc4b6d7acebe917fc0f490b22eefbc0f150e10ce5305c",
80
+ "snapshot/src/full_bandwidth_loop_tf/hf/train.py": "5767c6d879066068860b173a0b212ed82928ab42e5bbbd8c93bdd3cc032533fe",
81
+ "snapshot/src/full_bandwidth_loop_tf/input_write.py": "a7974576081355741c3188d38508f78da95db913347cd5a60396f98932d4b05b",
82
+ "snapshot/src/full_bandwidth_loop_tf/input_write_policy.py": "51c68dbc21d7a2fdd604669d97a464207cb5ad0200dfab3b0ef9ed80edfa723a",
83
+ "snapshot/src/full_bandwidth_loop_tf/kv_reuse.py": "7ac602c587d10b83768e067ec7c0cc358d924028ca76abff7209f411853074c2",
84
+ "snapshot/src/full_bandwidth_loop_tf/megatron/__init__.py": "82327efe57c404760d9b8cf9c083e85c9e526fc1c53e782d156adaa66b9b3f60",
85
+ "snapshot/src/full_bandwidth_loop_tf/megatron/adapter_decoder.py": "aea500799671712ebcd6683cb1fb0fccd648791d2ca103e9caa5050da787a542",
86
+ "snapshot/src/full_bandwidth_loop_tf/megatron/attention.py": "6835e2149abec4abb848b52406466a12d654cab2bebe19c129a679f64c82f91c",
87
+ "snapshot/src/full_bandwidth_loop_tf/megatron/builder.py": "3d8391ce8c65c721acd1e6fdefc07bdb94d888ffea042eb88b6fe10fdb702854",
88
+ "snapshot/src/full_bandwidth_loop_tf/megatron/config.py": "1c5801721a7a75de8bc65d138e0c0dbeee86d4113bb20938ab664e2672ba4839",
89
+ "snapshot/src/full_bandwidth_loop_tf/megatron/decoder.py": "bea4c2ff62dfebcbf08a156176cfe65f8d31c7df1fa24ea3c4991421d1e99a14",
90
+ "snapshot/src/full_bandwidth_loop_tf/megatron/model.py": "dee02d33cd9c8d389f71af71bf7e53d73ab5a886107335ca8b335bb57f0cb01e",
91
+ "snapshot/src/full_bandwidth_loop_tf/megatron/plugin.py": "ca8ee90c1e547ba7935c46fd2c2f2cd436940151c4915a4480eaa60243033728",
92
+ "snapshot/src/full_bandwidth_loop_tf/megatron/task.py": "c28c28caae4cc9e098d6a7a600acac970d6fc10928063cb4a1be5264106dfa92",
93
+ "snapshot/src/full_bandwidth_loop_tf/megatron/train.py": "4ffb371d68045560a0c48297fdb686592573e44dac826fbd4605639cd317f4b0",
94
+ "snapshot/src/full_bandwidth_loop_tf/policy.py": "a80e630571be5468a0ef0e1788cd7d6520e8dc158bcdfe6576bf47fe6917e2d2",
95
+ "snapshot/src/full_bandwidth_tf.egg-info/PKG-INFO": "f2d88943128cab5233cab07b7b805b4ee5e8aa96f9f073d010117c9644509e97",
96
+ "snapshot/src/full_bandwidth_tf.egg-info/SOURCES.txt": "4c70574587c35de3aa604039b72b77d1c2b6d66e25c5ed143988c991e6054b21",
97
+ "snapshot/src/full_bandwidth_tf.egg-info/dependency_links.txt": "01ba4719c80b6fe911b091a7c05124b64eeece964e09c058ef8f9805daca546b",
98
+ "snapshot/src/full_bandwidth_tf.egg-info/entry_points.txt": "44998f7c1bc5c962ad977370cdc1eace976e66b1c13c2b93e50bd274fef45eeb",
99
+ "snapshot/src/full_bandwidth_tf.egg-info/requires.txt": "a2520fc5016c6e2a199403891721d3044b08e0fc1af68eba8a0ce29b9080096e",
100
+ "snapshot/src/full_bandwidth_tf.egg-info/top_level.txt": "0212ad8598b39436853cd1881f4e815d69e4892925d6af74359f3e223eaae5aa",
101
+ "snapshot/src/full_bandwidth_tf/__init__.py": "73932d7b3105e983f86321784ff547181848cfde537630ab60e93b96c5c80729",
102
+ "snapshot/src/full_bandwidth_tf/hf/__init__.py": "6d9ae0e0360910223451d706e71477e258a4bea6ebaaa99287c77abc4b8a0130",
103
+ "snapshot/src/full_bandwidth_tf/hf/backbone.py": "da1062836060d9ea1580366ba4ecea04e409da1b5f23d328047040d44a5154f0",
104
+ "snapshot/src/full_bandwidth_tf/hf/config.py": "dabcc7e08f9d7026515d67393e0803f59c843598c59f790ec5934d6154cb8c27",
105
+ "snapshot/src/full_bandwidth_tf/hf/data.py": "9c9c3f53eae324c0c2acac63898b155f555f0c65191349c906d7530ebd0ab0dc",
106
+ "snapshot/src/full_bandwidth_tf/hf/diagnostics.py": "40dff542e9d801872b1e1bd2dca35bf5f39aa38dd1ece67f44c1b81aaf72cd92",
107
+ "snapshot/src/full_bandwidth_tf/hf/generate.py": "75fabd6e56c0d19de30a0743623277e7ef1f1c242fb069b6ec5acefe19d030dd",
108
+ "snapshot/src/full_bandwidth_tf/hf/inference.py": "9424da65353b2a55a2ecf179bcbf3e3d7c5fee735b6951cf8b356f4621dd7894",
109
+ "snapshot/src/full_bandwidth_tf/hf/model.py": "d9037d8b01b5157b2c591244293de5d14619c503afe424e6a754f53a358d171a",
110
+ "snapshot/src/full_bandwidth_tf/hf/plugin.py": "5f9cf723a71a2b1db9a3a03f395be964a4e175f2a60d493872c136fb2987035b",
111
+ "snapshot/src/full_bandwidth_tf/hf/task.py": "523d09295a271969a1b54369e6344be02850052c83f0135b3becb929b8d65139",
112
+ "snapshot/src/full_bandwidth_tf/hf/train.py": "3c83db8e6a91088fe00e781a69d218f2d0c4ae1317a8ecd86cbb33045d3002e2",
113
+ "snapshot/src/full_bandwidth_tf/inference.py": "6e100bc0ee8792bfe85b0042e25e24a04d433b133377da301ab678f6b6f26a5d",
114
+ "snapshot/src/full_bandwidth_tf/launch.py": "5f7fc124064b62effb6ea309c8bc7a15336c2a51d321505843cfe4a7ca8dfc50",
115
+ "snapshot/src/full_bandwidth_tf/megatron/__init__.py": "0ecf759960f5e1d412b3939b54394734df3ee54662fb73cbf621b9c773a8b3ef",
116
+ "snapshot/src/full_bandwidth_tf/megatron/builder.py": "770af80866b19ac6bf785c61658b874ebd14b5ebecbcc3699410ac8ca252b290",
117
+ "snapshot/src/full_bandwidth_tf/megatron/config.py": "1279e4ce26146a9879bce34eaeec2e7f53392a79cebf949d2942b70f9a1d2441",
118
+ "snapshot/src/full_bandwidth_tf/megatron/metrics.py": "bb7fb343d92ed6cd95a5bc091dfad26dfe94b9f4fb9c9dd22d05353f2ec62938",
119
+ "snapshot/src/full_bandwidth_tf/megatron/model.py": "1fddc5e768c0b7ef1ac2e14aa9fe7e2107eeadad4dddf8b5982f08a53691a428",
120
+ "snapshot/src/full_bandwidth_tf/megatron/plugin.py": "ddc601d23017d25423f67b309aadaba1272c0fc11a4fc53c755f754f4317cb0d",
121
+ "snapshot/src/full_bandwidth_tf/megatron/task.py": "99bc52999dbd72285252b855bd61709c3388dd2a3dcf47e52c643cc8c052ce67",
122
+ "snapshot/src/full_bandwidth_tf/megatron/train.py": "7da747ddd915c361b04efd18dbdd6afc99243dc866be0fbb71b596460f49e4d5",
123
+ "snapshot/src/full_bandwidth_tf/mtp.py": "77c6217551615384289fe676187414bff35a229f89032cb734a32d649a8c5cce",
124
+ "snapshot/src/full_bandwidth_tf/policy.py": "ccef928f79fd85b5b270f9d4c1c7a32b30151fb2e3751432109bd0406aa52d5a",
125
+ "snapshot/src/full_bandwidth_tf/validation.py": "935f19865a4e5b34962d21ef3c00d3c3614acd094bf2fcba579d0779eb8765ad",
126
+ "snapshot/src/hrm_text_tf/README.md": "26823181babe288b10168d06d5f37ce8b45a8fe49517b5492875a47b66c3affb",
127
+ "snapshot/src/hrm_text_tf/__init__.py": "620200ffcbd80720ba089f4ab4ab7e72ec4f6b173a39d1861afdf48402021fc3",
128
+ "snapshot/src/hrm_text_tf/check.py": "0959c3a0e2205b5e408246846ac2e7d7c01ac37d558901c1fba918aad49ccec3",
129
+ "snapshot/src/hrm_text_tf/hf/__init__.py": "70a45fbb9b72166003150ff69545dc8a143ca408fe26e6ed216b2381cb9359ac",
130
+ "snapshot/src/hrm_text_tf/hf/comparison.py": "abca6729c41be1fd23ff9f6d91612329d25f5db144f88ab61082647a0c2b43af",
131
+ "snapshot/src/hrm_text_tf/hf/config.py": "e92b8041331d5fea07d2ff5769bb3f43e6c664a827df4e7eee11972a8de4e4f6",
132
+ "snapshot/src/hrm_text_tf/hf/data.py": "7190165dc31c52f5e248b6d61b3be84b952c6ae45e758e37df1f6cbe1a3bdb9b",
133
+ "snapshot/src/hrm_text_tf/hf/lookahead.py": "81d4bbdfd717fa5d3bd4f769e56a6e1ea294980978a60de3aafa10193f8a9496",
134
+ "snapshot/src/hrm_text_tf/hf/lookahead_backbone.py": "6b9e596156a75458621863700d0310a0c67afc822a630b54c380294533a8b306",
135
+ "snapshot/src/hrm_text_tf/hf/model.py": "c0216a42520d758ed0ce49eec8dd862cbd2806578e96322f36508ec8d85c156c",
136
+ "snapshot/src/hrm_text_tf/hf/plugin.py": "db2a59aa0d0b0e1c6644d44c6cb1fc65123124b9683f4fb56b42a713c0fc85ed",
137
+ "snapshot/src/hrm_text_tf/hf/task.py": "32d0c615d455afbe51753c59c15452ec61370e97b560ae2d93c3fdce08961107",
138
+ "snapshot/src/hrm_text_tf/hf/train.py": "6625025b4d741b08641f71252d17871b8ce95c3b365d8f34e904542e0bca8b4e",
139
+ "snapshot/src/huginn_tf/__init__.py": "39d5e260fe7267a636c4e674a00972c727b74cd961e80cc9ae3af76fadf5d213",
140
+ "snapshot/src/huginn_tf/hf/__init__.py": "012202a3f5f7ad94fb75141e9e7acd572e736d35037396403b937071197a8f58",
141
+ "snapshot/src/huginn_tf/hf/config.py": "988917571eda3428ebbcb3b396dbca1040f78350bb6eb20d844d1a5573949307",
142
+ "snapshot/src/huginn_tf/hf/model.py": "e55e1b0b4755c517175818f74629a9b04c047547e95b09cf3408c49c31a337bd",
143
+ "snapshot/src/huginn_tf/hf/plugin.py": "f13fb4625549b47af55be1f1ef3200cfdebeed16be3d5fb0784d8eab14d41dcc",
144
+ "snapshot/src/huginn_tf/hf/task.py": "ca04d330852ba8c0c190be72d95c8802a65701fed9848ea3af8d608f7c85e6dc",
145
+ "snapshot/src/huginn_tf/hf/train.py": "a3f8facc58f3d13e835fa36829d8b8ee72123ac098604f1ce4be615e8743e8f6",
146
+ "snapshot/src/huginn_tf/megatron/__init__.py": "01a90198bdfa78ebf53ee3cf69078c7c6c11ccfc9dfdbbb840348c0453d884e3",
147
+ "snapshot/src/huginn_tf/megatron/builder.py": "a30735ff0ffe0df2340d8b0a351dd82792cd2ee0dda5488954f0e4dec3d954e2",
148
+ "snapshot/src/huginn_tf/megatron/config.py": "82731a0f18fc14825672c085d8febb395bb4196a60177453d852f8297ecd3f2f",
149
+ "snapshot/src/huginn_tf/megatron/model.py": "208d1ce0da3753171efe2cdd2aee01e882bf067c1135c4ac4ef7a1cce8adeee3",
150
+ "snapshot/src/huginn_tf/megatron/plugin.py": "7c1b7e54758ac416d5f4af1cc0f79c7ff98c292d1f5721c55246486ae33e5093",
151
+ "snapshot/src/huginn_tf/megatron/task.py": "43f8ed00f51e7bac2176c5a35060b4d12ad8a11f5b46dd3aaca690a599963a01",
152
+ "snapshot/src/huginn_tf/megatron/train.py": "d5636e052befdd38e4f971c8a7f38b7178cac783c6295e2ca787f8fa742fb5dd",
153
+ "snapshot/src/huginn_tf/policy.py": "05393214dbb2f9abdf0c1b317d5254d7888e9de52411f62c345cbe90fbb987ca",
154
+ "snapshot/src/huginn_tf/validation.py": "7dd329bbf0bef8dc287f2d337cb66cc479ba90cd7e138f5fcc25ed1881f10671",
155
+ "snapshot/src/lookahead_tf/__init__.py": "58894edf8fb480516d6667e25e88fe198f830b8de2cf9e6b71bd926137779a28",
156
+ "snapshot/src/lookahead_tf/hf/__init__.py": "75016b1cc8b2fae544c7fdbbfa5c7d0511217a63261223f0920fd6acf3be3bea",
157
+ "snapshot/src/lookahead_tf/hf/attention.py": "251eb59becae4b2d1796ea92cc10e12cd1f9aa3381704dc39b07ded4dd603637",
158
+ "snapshot/src/lookahead_tf/hf/attention_reference.py": "79c0db25b0667cc0f8128d224009ce85cc9021f7d0b36acb1e55654fdc631975",
159
+ "snapshot/src/lookahead_tf/hf/backbone.py": "3da59b4c4a1bd2df0b79ec4d1c70be81076c4aa9f38eda1d3381f2f7b3fbeecc",
160
+ "snapshot/src/lookahead_tf/hf/config.py": "67bf0756a3eee7e02086b37587e14bebf022651e49faabd6d4e592405d9637b3",
161
+ "snapshot/src/lookahead_tf/hf/decoding.py": "978a30f2af155bc1759cf9eafc29f350c4bf3e7506ae4e40666e89cd233292fc",
162
+ "snapshot/src/lookahead_tf/hf/diagnostics.py": "1942539a5947dcbe3c1ca15d7cd8558c0538453a0444c9bd1fa9646c1ab1cd7a",
163
+ "snapshot/src/lookahead_tf/hf/inference.py": "1a4fa723597404619d51b386be18d6d175813aff500b99fc327ca3174cca4588",
164
+ "snapshot/src/lookahead_tf/hf/losses.py": "d16a0bc5538fccbd51e5bebfc18668a6c06be0a613f51368353e76510135dcb8",
165
+ "snapshot/src/lookahead_tf/hf/model.py": "1d8c639494c6068301a5322e0142e45cc37a7755f0c36b63132cfa2e1a4e64ae",
166
+ "snapshot/src/lookahead_tf/hf/paper_backbone.py": "6147f1b50a5e5d30844973537a455784ed60385cf7cfdba54029043a93a13ce5",
167
+ "snapshot/src/lookahead_tf/hf/plugin.py": "85576c47247a5267af087bb863a3b3f5b9c8743eb3dc6d8138000edd2f0040f5",
168
+ "snapshot/src/lookahead_tf/hf/refiner.py": "d28b8c5f5c92cf5f69abbce8b2124c98ebc51fbc720be0d8578f25f430029b66",
169
+ "snapshot/src/lookahead_tf/hf/state.py": "0005763d132db35d9c181165b3bd38116a6e265eb4efbbcc42fbc4c498204c64",
170
+ "snapshot/src/lookahead_tf/hf/task.py": "5ea98496e38a614504a31d41aa33279fd875b8377133fbabe3d75713cf10ccb3",
171
+ "snapshot/src/lookahead_tf/hf/train.py": "5d63387f67c4a8bdeb3b641e2736c5fd1825cf66ba2c3069fe79320f39bd5670",
172
+ "snapshot/src/lookahead_tf/hf/training.py": "412ae9a9b75cbb7152c35f77e7b5c3e6bcdbe2a05ea582cd1342415036546aa5",
173
+ "snapshot/src/lookahead_tf/initialization.py": "79a0d1a346141b123bc4d21599988b0cf13ec136eda82f641b70d2f06cf8c6c5",
174
+ "snapshot/src/lookahead_tf/megatron/__init__.py": "83ccc5a2293c077268c8bf855672b1020caf8d1396c338f8570af629a4456a1e",
175
+ "snapshot/src/lookahead_tf/megatron/builder.py": "a605d8b146f6a73f225a87bf981294a5ee6d89ac11b7ab00221b8e68b17c22ee",
176
+ "snapshot/src/lookahead_tf/megatron/checkpoint.py": "5ba116561ef87094b7c208d884d1f98af030d8df47d11965231528d5816ab2c4",
177
+ "snapshot/src/lookahead_tf/megatron/config.py": "2887c4303fdb0c5076940442bb9d0221ea0dc119214fceae5f572bde15af4f56",
178
+ "snapshot/src/lookahead_tf/megatron/model.py": "adbf72161cc40e8caab9a53d700e3661b24552dc3edc25921143350b44e36cd8",
179
+ "snapshot/src/lookahead_tf/megatron/plugin.py": "c2a6e985600641a77e173ecbfceb561c9c23aa98e3b2ed7509fb75f59907fedf",
180
+ "snapshot/src/lookahead_tf/megatron/task.py": "2e6d660e454ef8153e3c73cb64a656e45232fe60ffaed4fb39c577c6f429a412",
181
+ "snapshot/src/lookahead_tf/megatron/train.py": "02a71c985dc8ff8a63631dbd9baa9af28fb8ab838def1a9df799b821bcf76551",
182
+ "snapshot/src/lookahead_tf/policy.py": "7c5ea760e2243392460b1417b102365d70906b3c151e1e2355d5f3bb53789954",
183
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/README.md": "742136ba9b9c6d8e020a6721469c833b95f3b33551957333ba9bcc83864ecb25",
184
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/__init__.py": "1d8fcf503f0d78a23e448155572aa57201c357b38e7cf1b563bd064b64ce197c",
185
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/layers.py": "6ff5451060c7ab8113b004021d3e59d40ad41556e0dafabf564dca8717d315f5",
186
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/megatron/__init__.py": "410bb1030d04a7a8bc692b422e5e52466699d96123581e05c8dcbb4d7814320b",
187
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/megatron/layer_spec.py": "691ab807a84b9023cab4a28a296695f6fb6cbffcc3e1d3b124760d7c0f0f3833",
188
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/profile.py": "794fccdb5767e1b8325b466893fe087a2638e206d2b6c692af5e21b82c34653b",
189
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/torch/__init__.py": "dda23e33011a0dc06211a9a093c92a341feb82e870819b50ad719487967f4626",
190
+ "snapshot/src/loop_transformer_backbones/LoopTFDR/torch/block.py": "3e178c8c11aadc72cdbf215f8bd8717b93a0157147b236989480376317e183d2",
191
+ "snapshot/src/loop_transformer_backbones/__init__.py": "8517ac48322900e373518ef0c557229f02a9558a71003820f5c9b21ef3b5d587",
192
+ "snapshot/src/loop_transformer_backbones/attention.py": "775e6571e467fdffdde3815d02c647edf675611905d5de9539fdd661b4131492",
193
+ "snapshot/src/loop_transformer_backbones/hrm_text/__init__.py": "38a673898d1b96a9137eb6f369752e734575a63a95d722a1511d5f9855c8d765",
194
+ "snapshot/src/loop_transformer_backbones/hrm_text/block.py": "113b5a249462682d286e1833121266185aa29470a16c243bff6347e5539ad93c",
195
+ "snapshot/src/loop_transformer_backbones/packing.py": "fceb4a6017017a428945bca322d8a684c46e439578c5eebeeed497c7510dbdda",
196
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/__init__.py": "3458d444e0795105ca9c40a0136be6616eed25d0511de4ea4bd73efb2204d57c",
197
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/initialization.py": "530597dccacf6669057b75f52ee7b8654b7722e9a32295d27443bc56f205ed9d",
198
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/layers.py": "882edb527b62820e794ccb5d8617d76934e41a0725f25fdb0463d8edf0c008c0",
199
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/megatron/__init__.py": "3458d444e0795105ca9c40a0136be6616eed25d0511de4ea4bd73efb2204d57c",
200
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/megatron/initialization.py": "3eacb55bc58ce6263f40e9b5ff5aaa4c47fa69b045c2ffe0bc00788386f5ec0f",
201
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/megatron/layer_spec.py": "a1bb0f4b424664acd33069c4efc1fe67287cf2624c7425691df85aef348fd75b",
202
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/profile.py": "6c36cd38f3a8d3445a2d1a3d53da013f771e3a3474cca5e4abe10a6f5df64f08",
203
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/torch/__init__.py": "3458d444e0795105ca9c40a0136be6616eed25d0511de4ea4bd73efb2204d57c",
204
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/torch/backbone.py": "dcace7df22bf1498ee7436ee3a6fc08f74d790b63150142da2572708f6b13602",
205
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/torch/block.py": "fa6667637901dd3581d486c5d02cdf42b33f56817e17dba26cd165841e41643b",
206
+ "snapshot/src/loop_transformer_backbones/paper2604_21106/validation.py": "4cdb33b4e1d359bf4d588903249f26c1eb21f647108f93966f43d30121314aba",
207
+ "snapshot/src/loop_transformer_backbones/paper2608_08888/README.md": "8f39f91e5d2c9797fdf95da79b74f2c433b276a94a3ee3c62534595063f7ab78",
208
+ "snapshot/src/loop_transformer_backbones/paper2608_08888/__init__.py": "74583d5f333a24f0914acd3fe8ed8c8549d026f0cd73f9ed469e14749f91ead2",
209
+ "snapshot/src/loop_transformer_backbones/paper2608_08888/backbone.py": "cbcb042c08de4ba7bf9128cda9cc295e39f9f63ff1e416c40f621052c97c02e9",
210
+ "snapshot/src/loop_transformer_backbones/shared/__init__.py": "928c302cd7510b6cc04eeb0c955b19f31f5fdb5136d77748206673a2aae185d5",
211
+ "snapshot/src/loop_transformer_backbones/shared/attention.py": "d4b15037eb5011abb9777e8c7fa274fb2f83baf0fcdc916543878d573309ae7a",
212
+ "snapshot/src/loop_transformer_backbones/shared/rope.py": "1515180bcc9ba71bd31b2d514bd8402d0a1a42931cf76b27a923051a35ad1cb2",
213
+ "snapshot/src/loop_transformer_backends/README.md": "6fade4fd6073b2c39109bb6a5e8f85737e5ebd13ab8c0b9f123859bc13090521",
214
+ "snapshot/src/loop_transformer_backends/__init__.py": "1f97d268753cd796d5c663bf7a6e9e709f193efef88ecaa1fb463fd26ce8ab29",
215
+ "snapshot/src/loop_transformer_backends/backbone_losses.py": "7ebc1618d504cd2370c1792e16fe404a0b68950c320bef250802142001e1f90e",
216
+ "snapshot/src/loop_transformer_backends/benchmarks/__init__.py": "112d1ace8221c58c5798e5b40ad52c2fddf3f4c450016cac0fb8948b38750502",
217
+ "snapshot/src/loop_transformer_backends/benchmarks/cli.py": "bb1bbde391ea814550dcdf6d69ee63b8daa5e1e9d824e0d08046e688ae70d4be",
218
+ "snapshot/src/loop_transformer_backends/benchmarks/code_scoring.py": "20c635f8d4739b788d82ed427798b2163c94b7e58133fa70d24889207213dda7",
219
+ "snapshot/src/loop_transformer_backends/benchmarks/contracts.py": "3a16b7684fbc78554c0c5f6ad83689da00dd8fb2e7d4f0322090d65555dfeac1",
220
+ "snapshot/src/loop_transformer_backends/benchmarks/harness.py": "77f800da638dfb295c2bc47b064f75f3e304e6830f06e8cb2cc121de58a95bb2",
221
+ "snapshot/src/loop_transformer_backends/benchmarks/ifm.py": "83cdfa5444f9efde75234ff0c0e55ccd76225d61bf213159845a3a7cf0ee72ad",
222
+ "snapshot/src/loop_transformer_backends/benchmarks/ifm_data.py": "48fac62274cf48f3320f4fa2917c252d7b057db376646ec810c3de02d4319332",
223
+ "snapshot/src/loop_transformer_backends/benchmarks/inference.py": "3a18955200b4d6ecf96022a3fc606bdd540bc7d627d4760f70025206fc9ced0e",
224
+ "snapshot/src/loop_transformer_backends/benchmarks/models.py": "74813733654e6b8186ca0470e5f538422ee65405f3978bd533e4ec1bff8ede59",
225
+ "snapshot/src/loop_transformer_backends/benchmarks/native.py": "b8703f179e0ef5ed228a910aa4d7846b2b74fd11a8ee6f2cf42abf40ca7c6588",
226
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/__init__.py": "bc7b4f54523aebc6bbb1493799eb8c044542187174c50adc41b07234b8e86558",
227
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/data.py": "f04ddc0f3d0923d2d4943cd612d16245e4b29ff4e9039abcedf283b2e07f3951",
228
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/models.py": "a30f13226af60138571be0fc8b6059485e91f7c6487f13da0830dc6218b49e67",
229
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/prefix_query.py": "c6280a79d5adf0eb69e6eb0c67602b7eb91c0b030b23a6ac7a1e553435e752ee",
230
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/runner.py": "c0afca509646e7d2ebaf23f076d1990ed91df6ae6787e00453a04eff224ca69b",
231
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/scoring.py": "18655cf51a3741f491862903c77c69bd4eaeceb474c6f9f13816e1f7a86fb522",
232
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/vendor/README.md": "454cbe8b8813b00922cfb121bc839813b3bbabc2b7d3dae09942e70eddc80f75",
233
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/vendor/__init__.py": "17ca169fe2554e86a286e9b02c93d5b500b2790e3c61a333bea576189dbd1791",
234
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/vendor/core_eval.py": "d2fd43fd18b0764a15c05110238ff92903cf15dc0bbb33222f6d83b059db2b65",
235
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/vendor/prepare_owned.py": "d855ab38b32f4510384dd14a7f4da433b8f96d7132811a08e436758ffe63b166",
236
+ "snapshot/src/loop_transformer_backends/benchmarks/paper2604/vendor/prepare_saunshi.py": "63bf92ea02014c182861cf6b0ace09c99e06a57d83db46d9885ce8c3b6763ec9",
237
+ "snapshot/src/loop_transformer_backends/benchmarks/receipts.py": "5a908cfa188b509d98b7de5af71fb979a03577dde3cd2d34d8d0c76b8803f37e",
238
+ "snapshot/src/loop_transformer_backends/benchmarks/task_sources.py": "67a520c0ad121949eabbfa1d8c1702df3b47960e758279958cd6f64f20c4447e",
239
+ "snapshot/src/loop_transformer_backends/causal_hrm_text/__init__.py": "66b9d23e61abd58ea554b8e96c6a333157f37051b20fb6ea8991d69fbc2bc8aa",
240
+ "snapshot/src/loop_transformer_backends/causal_hrm_text/check.py": "c6e76a3613916cf442a8879fdbd9139b4ad68a6dc19c008e888ae039d3a942ea",
241
+ "snapshot/src/loop_transformer_backends/causal_hrm_text/config.py": "17aaebd942ef355247fcffe340ef55b7d348ca8313d3dd58b26e7c3a63c79159",
242
+ "snapshot/src/loop_transformer_backends/causal_hrm_text/data.py": "1ca1a7944d4baa663f8c7819b1c844b1e6aaf409a78067a48142a2e77834dbed",
243
+ "snapshot/src/loop_transformer_backends/causal_hrm_text/model.py": "9aa24f4dfbdd53c76ebf3fd30f5dc020b75f0a08c772e7b6fcdad77ffcf330fc",
244
+ "snapshot/src/loop_transformer_backends/causal_hrm_text/task.py": "ca3ce7017467da482705e7901cf35971958e28ebd7e815981fde5190e8cc96c7",
245
+ "snapshot/src/loop_transformer_backends/compute_budget.py": "61ebe853d3cad27ecf1b5fa9c26dbde45993d80566899bd50664e59246312fd2",
246
+ "snapshot/src/loop_transformer_backends/defaults.py": "52f9aa466cfc3d6dc5919d43a9b52f07ad9ff68d84388328a3debb57ad0bac18",
247
+ "snapshot/src/loop_transformer_backends/evaluation_queue/__init__.py": "59585dc8311358fc00f265e56906d73ec9c20a9acce497a22d37fc66d5566c3b",
248
+ "snapshot/src/loop_transformer_backends/evaluation_queue/campaign.py": "7123c21610535993e21bde11c42773bddcb12379b83c06545c0aa12f871d41c1",
249
+ "snapshot/src/loop_transformer_backends/evaluation_queue/catalog.py": "922c1b8af357358a24b900ccbc88144fa3a2076d119854874fd8d514bc57ad8c",
250
+ "snapshot/src/loop_transformer_backends/evaluation_queue/dispatcher.py": "d5404c376ebf998bb6972583e1fa903404e944cd51230898e02fb4b23b62e78d",
251
+ "snapshot/src/loop_transformer_backends/evaluation_queue/harvest.py": "2c2cd8f1801c254f7ad308b5308a357c221756f8c9a1706fd127d4465d3dec30",
252
+ "snapshot/src/loop_transformer_backends/evaluation_queue/naming.py": "6442e3e08a4ec87dc2c6d92d543ae6187b6f26ccf135701c613ce7f26a2fc2e2",
253
+ "snapshot/src/loop_transformer_backends/evaluation_queue/protocol.py": "d90ad08fa56a45842cb1eaf34ed6db14c48659dd98862d4e4c7b646b247fc3bf",
254
+ "snapshot/src/loop_transformer_backends/evaluation_queue/split.py": "3ad95ea1be5bb8501fa2537ba995a8d8a0f9194a4c78d9ada23d3585d38e7691",
255
+ "snapshot/src/loop_transformer_backends/evaluation_queue/worker.py": "d69c7861c93a344823a7e7aa3e8d07c1c490b623a86eed43c8788e5ceca41dc2",
256
+ "snapshot/src/loop_transformer_backends/loop_budget.py": "5fa278a01178926f69ecf41e497eba0a0be40b2742ae7bbbe591084fb5af0ec5",
257
+ "snapshot/src/loop_transformer_backends/loop_horizon.py": "4b33fff6f8a7db10d3cc14ba4cdf47dbee343d2a8f6d3d912bf948ddba427c6b",
258
+ "snapshot/src/loop_transformer_backends/loop_search.py": "9495435ea0c6b36fb780a9e3f8c1c26f74bf580daf28f446ceddce67dd0e161d",
259
+ "snapshot/src/loop_transformer_backends/megatron/__init__.py": "e34071777fb85eec3ff4c785ec243229259c9f74c1cbd0fc5281285653212838",
260
+ "snapshot/src/loop_transformer_backends/megatron/compat.py": "b15db2b52c81221878ea3eb4f86d2829cb9d13da52a5bbdef26bdfe64af350e8",
261
+ "snapshot/src/loop_transformer_backends/megatron/contracts.py": "05f4528ef5c07371caea28cfe563b8240f71a051d079e9d5ae11e041ef3447be",
262
+ "snapshot/src/loop_transformer_backends/megatron/data/__init__.py": "95095cdda3d00fff34919f2058f5eaadcdc458944e43d7d21cc85adc6b75f2c8",
263
+ "snapshot/src/loop_transformer_backends/megatron/data/batches.py": "0149909bfc246aceb268f092bf84e856bcedf55be6478f8a58395cbf50bd27e1",
264
+ "snapshot/src/loop_transformer_backends/megatron/data/datasets.py": "117405f41d47870028852a8fbd8de4a452b18c357f61efdd8c4dd2f48a5ac6a0",
265
+ "snapshot/src/loop_transformer_backends/megatron/data/online.py": "152ed6230fdaf4e5542fb3ff40045887d07c21ef47db820fa539d5dfc276f463",
266
+ "snapshot/src/loop_transformer_backends/megatron/data/prefetch.py": "ad239e2a7760a7cc590afd446160fc1e808dc4cdb93d011b071427c13a571ada",
267
+ "snapshot/src/loop_transformer_backends/megatron/data/raw_cache.py": "31aab96d544ecccea5168eb46d5dce218fb5ae039bc8ae771e315cf67dbbe619",
268
+ "snapshot/src/loop_transformer_backends/megatron/data/streaming.py": "86c472554205f1ce6317833fa43410a1ce5a7327599ec711ed67052821e90bea",
269
+ "snapshot/src/loop_transformer_backends/megatron/data/validation_cache.py": "3b40979c919faf2b5e0e85f62db97ea4dc544431134f00874158cf7849c256ba",
270
+ "snapshot/src/loop_transformer_backends/megatron/evaluation.py": "4fa012b45c8014ef067f2669231789178be63c6c0084c7d687203c3c66526f79",
271
+ "snapshot/src/loop_transformer_backends/megatron/losses.py": "a258b7833d072537e3b2100b06709ec8c19855e2b71a82eb799190006cb6aeef",
272
+ "snapshot/src/loop_transformer_backends/megatron/metrics.py": "174f60c0fa806472884f4ea3e4cfde08e9549acc04114fef3c84e25ebc9d7441",
273
+ "snapshot/src/loop_transformer_backends/megatron/recipe/__init__.py": "c4e2d48cb2beca7b37ba30496be169f11092e80cdd65eb8ec6b318a76dd1f857",
274
+ "snapshot/src/loop_transformer_backends/megatron/recipe/legacy_normuon_wsd.py": "ba41ca5350b032f1e44c62c9d254d9ce500bda1a15ccddc59dfa162431fbc9c5",
275
+ "snapshot/src/loop_transformer_backends/megatron/recipe/paper2604_21106.py": "69df1d414019e2352021021b7926aea199b7c95a53dc075b10edeb2c2f098fc2",
276
+ "snapshot/src/loop_transformer_backends/megatron/recipe/paper2608_08888.py": "6a706d05c2db4b7dae1fcf51408e2d466c5e58cc957d98512d65b2f3d107088f",
277
+ "snapshot/src/loop_transformer_backends/megatron/recipe/shared_configuration.py": "1095f8ed7a40ed5f037f764fb5750223194792898f578e3e455b4f50bf7af907",
278
+ "snapshot/src/loop_transformer_backends/megatron/recipe/shared_schedule.py": "5ad080fc1b52a66284d3eab3f15d709ce23c752c4528ae65b648cfa56bc8bf15",
279
+ "snapshot/src/loop_transformer_backends/megatron/registry.py": "6c74d9e843b9d97c4ecdbdc678af2046aa8d1e5062190156864cf4c05ad472ac",
280
+ "snapshot/src/loop_transformer_backends/megatron/runtime.py": "2fafa8ea432b098c50a457b30183f4128f695d6f0bac080e14093e2cee0e7b1d",
281
+ "snapshot/src/loop_transformer_backends/megatron/validation.py": "b8965ab1459e023025447c5b56e3e1f3eaa85c430e3fd1845257e04d3e451284",
282
+ "snapshot/src/loop_transformer_backends/megatron/wandb_hashing.py": "430208baf0399ae9d40b0211494802abb33a1f60845dad69e62a1310d3d09261",
283
+ "snapshot/src/loop_transformer_backends/recipes.py": "7eb45d66610b5317fbe7d01831e7cc02529ad458727ab29e9fa6ba29f70e74a4",
284
+ "snapshot/src/loop_transformer_backends/tokenizers/__init__.py": "c28a006de78276efa4110498c7c50b729469487670a80cd3da2b8895447a1ce6",
285
+ "snapshot/src/loop_transformer_backends/tokenizers/paper2604_21106.py": "3a3df8031d26bd9032c3b508d9990168e1b15c2bf8a1c4703ccfdb6ae7cb5e73",
286
+ "snapshot/src/loop_transformer_backends/torch_distributed/__init__.py": "c759ae93f0dc7bd282165417a46370c27ef61fcbd9ee96ba33843654ea67ff26",
287
+ "snapshot/src/loop_transformer_backends/torch_distributed/checkpointing.py": "6351d720d61a55b34a2f551206803db584c719852fb28d8559ce6b604dfb8535",
288
+ "snapshot/src/loop_transformer_backends/torch_distributed/config.py": "658470133fb34e35ecbdccc5fd6f4cf93b03df20fbed34a65728290a1ae7d521",
289
+ "snapshot/src/loop_transformer_backends/torch_distributed/contracts.py": "bf4de3792ea1664b61e7ca54d7cfa20476d542a07abe14c454a9e0b278de1264",
290
+ "snapshot/src/loop_transformer_backends/torch_distributed/data.py": "0633000a1e98c0469a6503c18bc0ca0b022ed4672e4d1e1cf06d07582322bae9",
291
+ "snapshot/src/loop_transformer_backends/torch_distributed/distributed.py": "ca634bf0f658fb8a9b635416c45dd75618a7d92bad9eed9dde02643c26f5615a",
292
+ "snapshot/src/loop_transformer_backends/torch_distributed/muonh.py": "c5fc541a0b25ded840dc31aa2769c0b72122ef46c1019162d754c7240bff2d52",
293
+ "snapshot/src/loop_transformer_backends/torch_distributed/optim.py": "4cccf133f01c65ba9821e25f5bef8530eb8aca96c29846e4985c8f09088b59ef",
294
+ "snapshot/src/loop_transformer_backends/torch_distributed/reductions.py": "42db3763bcb69352a9a4c592c2da0135300230692f741a4e50a8468e78cafc19",
295
+ "snapshot/src/loop_transformer_backends/torch_distributed/registry.py": "04297374a0f6747ca235193642ca2bbd90df1375a3c119be4496f6efe2832198",
296
+ "snapshot/src/loop_transformer_backends/torch_distributed/runtime.py": "b93940e3564e442cd65f2a8235e35f1350aac96d805703f0f86360f5756c8a41",
297
+ "snapshot/src/loop_transformer_backends/torch_distributed/schedules.py": "cfcf6e60e5ca17eb105e351535ff87ae85440fe0d78beba6dfc03799b9345379",
298
+ "snapshot/src/ouro_tf/__init__.py": "a73fdac27f075f98269937ecaba86e6c5e3b4bf36764f1350a6b5efddd25081a",
299
+ "snapshot/src/ouro_tf/hf/__init__.py": "b1c80c4f72cca2e9870ff2cfb717923ab2e29365dd4e91f3b484a8e9eebbf432",
300
+ "snapshot/src/ouro_tf/hf/backbone.py": "4f7cbaa3bbadde86c7f0a1a624f11f887247939e428e378a08ed00913ddc9c7b",
301
+ "snapshot/src/ouro_tf/hf/config.py": "40f640886aa8dc08995d4c106b6bfa2458f1b2ede8a34607ec73ab7ab6794c66",
302
+ "snapshot/src/ouro_tf/hf/model.py": "3bdc66c59a83e2d02349d78504f54653de3d76a9655bcda9e7b0c5ed5f662e73",
303
+ "snapshot/src/ouro_tf/hf/paper_backbone.py": "7ca0d348ace2d4476f840e7a35ae2db6e3ecf8889996dea50fda80060a5ab447",
304
+ "snapshot/src/ouro_tf/hf/plugin.py": "e63e830f2dbbbcc21ccd0d0373634e8e1a8a4da9d9fb5ec92e25621c8ac88c28",
305
+ "snapshot/src/ouro_tf/hf/task.py": "4f5ee8a7adc5d227e9ab717e7f930990dfbc86ab56cdb03bad7b55e1fdbe6b62",
306
+ "snapshot/src/ouro_tf/hf/train.py": "37e775223efd3a87ccf85db312b0e729c793afe42652702acd8a84070bfbe0be",
307
+ "snapshot/src/ouro_tf/megatron/__init__.py": "33608673413945f504d41a35eec819b32679d95b5440c2f5c4040211b629dd8d",
308
+ "snapshot/src/ouro_tf/megatron/builder.py": "903ff9c341ac38325465c16edc893cec3eac77546fb2336e5f552cc494fead55",
309
+ "snapshot/src/ouro_tf/megatron/config.py": "42400d266ff4f931ed714631c7131a38e163dedb58c88079413e066df5b2ec02",
310
+ "snapshot/src/ouro_tf/megatron/model.py": "a47f9b0323057c2051d7ba3661eab6fcb66ff8299368287d68b38efffb07e01d",
311
+ "snapshot/src/ouro_tf/megatron/plugin.py": "f503136dcc0f4dc9149b342e63c1294e7f289836630c35a83ccdeff056590127",
312
+ "snapshot/src/ouro_tf/megatron/task.py": "357d4dfbe54658408237ab4b46b117a70a0cddf9a5139db3cb5e0c9c8220bc2a",
313
+ "snapshot/src/ouro_tf/megatron/train.py": "094fb39647af251e9064241e9e3cfdf5eead1108e57de00c4f8e9f90da4a501f",
314
+ "snapshot/src/ouro_tf/policy.py": "431640e74eeadb5061080dc9e8773856d51642c48b77493223e50991137427b5"
315
+ },
316
+ "training_revision": {
317
+ "branch": "main",
318
+ "commit": "3167faf2170a85bf95755087d9bfd0e5515186e6",
319
+ "worktree": "dirty"
320
+ }
321
+ },
322
+ "tensor_count": 170,
323
+ "tensor_dtypes": {
324
+ "torch.float32": 170
325
+ },
326
+ "tokenizer": {
327
+ "active_vocab_size": 32001,
328
+ "path": "tokenizer"
329
+ },
330
+ "training": {
331
+ "final_checkpoint": true,
332
+ "global_batch_tokens": 32768,
333
+ "input_tokens": 1000079360,
334
+ "iteration": 30520,
335
+ "planned_input_tokens": 1000079360,
336
+ "seed": 42,
337
+ "sequence_length": 2048,
338
+ "train_iters": 30520
339
+ },
340
+ "training_run": "Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1",
341
+ "training_tokens": 1000079360
342
+ }
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/experiment.yaml ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ architecture:
2
+ name: loop_budget
3
+ options: {}
4
+ model:
5
+ model_type: full_bandwidth
6
+ pretrained_model_name_or_path: null
7
+ revision: null
8
+ trust_remote_code: false
9
+ attn_implementation: sdpa
10
+ vocab_size: 32001
11
+ hidden_size: 320
12
+ intermediate_size: 1664
13
+ num_hidden_layers: 20
14
+ num_attention_heads: 5
15
+ num_key_value_heads: 5
16
+ max_position_embeddings: 2048
17
+ paper_sliding_window: 2048
18
+ paper_full_attention_interval: 6
19
+ rope_theta: 10000.0
20
+ rms_norm_eps: 1.0e-06
21
+ initializer_range: 0.02
22
+ attention_dropout: 0.0
23
+ tie_word_embeddings: false
24
+ gradient_checkpointing: true
25
+ hf_config_overrides: {}
26
+ backbone_profile: paper2604_vanilla
27
+ padded_vocab_size: 32064
28
+ paper_allow_ffn_override: true
29
+ budget:
30
+ kind: huginn
31
+ inner_loops: 3
32
+ outer_passes: 3
33
+ prelude: 5
34
+ core: 10
35
+ coda: 5
36
+ backprop_steps: 3
37
+ ouro_objective: stage1
38
+ entropy_beta: 0.1
39
+ feedback_loss_weight: 1.0
40
+ normalize_outer_loss: true
41
+ target_parameters: 50000000
42
+ parameter_tolerance: 0.01
43
+ lookahead_refiner_intermediate_size: null
44
+ lookahead_refiner_layers: 4
45
+ lookahead_draft_loss_weight: 0.0
46
+ lookahead_future_loss_weight: 0.0
47
+ lookahead_auxiliary_loss_passes: final
48
+ lookahead_auxiliary_loss_weighting: sum
49
+ huginn_intermediate_size: null
50
+ compute:
51
+ mode: block_matched
52
+ receipt: null
53
+ target_flops: null
54
+ tolerance: 0.01
55
+ target_blocks_per_batch: 120
56
+ block_shortfall_allowance: 0
57
+ fixed_total_tokens: 1000079360
58
+ fixed_global_batch_tokens: 32768
59
+ fixed_sequence_length: 2048
60
+ horizon: null
61
+ optimizer:
62
+ recipe: paper2604_21106
63
+ recipe_overrides:
64
+ optimizer.matrix_lr: 0.014
65
+ name: muonh
66
+ matrix_lr: 0.014
67
+ scalar_lr: null
68
+ matrix_weight_decay: 0.0
69
+ scalar_weight_decay: 0.0
70
+ embedding_lr: 0.3
71
+ unembedding_lr: 0.004
72
+ norm_lr: 0.005
73
+ other_lr: 0.005
74
+ reference_width: 640
75
+ reference_tokens: 983315200
76
+ reference_batch_tokens: 262144
77
+ betas:
78
+ - 0.8
79
+ - 0.95
80
+ eps: 1.0e-10
81
+ normuon_momentum: 0.95
82
+ normuon_beta2: 0.95
83
+ normuon_nesterov: true
84
+ normuon_adjust_lr: spectral_norm
85
+ max_grad_norm: null
86
+ schedule:
87
+ kind: linear
88
+ warmup_steps: 0
89
+ cooldown_fraction: 0.0
90
+ cooldown_style: linear
91
+ final_lr_fraction: 0.1
92
+ z_loss_coefficient: 0.0
93
+ adamc_weight_decay: false
94
+ data:
95
+ train_bin: training-data-not-included/train.uint32
96
+ val_bin: training-data-not-included/validation.uint32
97
+ dtype: uint32
98
+ sequence_length: 2048
99
+ prepacked_next_token: true
100
+ eod_token_id: null
101
+ shuffle: false
102
+ seed: 42
103
+ training:
104
+ output_dir: inference-only
105
+ total_tokens: 1000079360
106
+ global_batch_tokens: 32768
107
+ micro_batch_size: 2
108
+ strategy: ddp
109
+ mixed_precision: bf16
110
+ log_every: 10
111
+ eval_every: 3052
112
+ eval_batches: 64
113
+ save_every: 6104
114
+ seed: 42
115
+ resume: null
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/manifest.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "checkpoint_iteration": 30520,
3
+ "files": {
4
+ "README.md": {
5
+ "sha256": "42b6692349dcbdc5018514b27f2e2ade7a21f5eea76ee1302687e8d033024453",
6
+ "size": 954
7
+ },
8
+ "config.json": {
9
+ "sha256": "6057b2212e1d8404194bb07444326346bcdfac035bb3f6d48f34baa9f70cab0c",
10
+ "size": 34881
11
+ },
12
+ "experiment.yaml": {
13
+ "sha256": "927fb814482243b95cb64c0d40931910cd995fc2655102f5ba99f0c638db2f20",
14
+ "size": 2692
15
+ },
16
+ "model.safetensors": {
17
+ "sha256": "f97b125cc6c7e2d436294796292abeb0cc117d02cf7f56a6a7eb6f50682c880e",
18
+ "size": 201763568
19
+ },
20
+ "tokenizer/special_tokens_map.json": {
21
+ "sha256": "99e1ac71ff7f0e3e60e45169cdfd7e8540b8c234e7146100e60a02c4751553ee",
22
+ "size": 134
23
+ },
24
+ "tokenizer/tokenizer.json": {
25
+ "sha256": "60c0314d888788ac33de8cb7262be527cffb0af30f0516cc3dd16b2b378bc6cb",
26
+ "size": 3619206
27
+ },
28
+ "tokenizer/tokenizer_config.json": {
29
+ "sha256": "77f3faab79c4008cb57c4baab3c75b69ef088c5fb15e8f7af59471fe37abc001",
30
+ "size": 988
31
+ }
32
+ },
33
+ "path_in_repo": "Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520",
34
+ "repo_id": "Siyuc/loopTF-workspace",
35
+ "repo_type": "model",
36
+ "schema_version": 1
37
+ }
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f97b125cc6c7e2d436294796292abeb0cc117d02cf7f56a6a7eb6f50682c880e
3
+ size 201763568
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/tokenizer/special_tokens_map.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|assistant_end|>"
4
+ ],
5
+ "bos_token": "<s>",
6
+ "eos_token": "</s>",
7
+ "unk_token": "<unk>"
8
+ }
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/tokenizer/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
Huginn-K3-P5-R10-C5-T3-seed42-50m-1b-BS32768-MatrixLR1/step_30520/tokenizer/tokenizer_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "0": {
4
+ "content": "<unk>",
5
+ "lstrip": false,
6
+ "normalized": true,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "1": {
12
+ "content": "<s>",
13
+ "lstrip": false,
14
+ "normalized": true,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "2": {
20
+ "content": "</s>",
21
+ "lstrip": false,
22
+ "normalized": true,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ },
27
+ "32000": {
28
+ "content": "<|assistant_end|>",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false,
33
+ "special": true
34
+ }
35
+ },
36
+ "additional_special_tokens": [
37
+ "<|assistant_end|>"
38
+ ],
39
+ "bos_token": "<s>",
40
+ "clean_up_tokenization_spaces": false,
41
+ "eos_token": "</s>",
42
+ "model_max_length": 1000000000000000019884624838656,
43
+ "tokenizer_class": "PreTrainedTokenizerFast",
44
+ "unk_token": "<unk>"
45
+ }