Upload Think.Unbounded-d32-v2mix-cont-nanochat-default-datamatch-v1 step 160
Browse files
experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-nanochat-default-datamatch-v1/checkpoints/meta_000160.json
ADDED
|
@@ -0,0 +1,250 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"step": 160,
|
| 3 |
+
"training_complete": true,
|
| 4 |
+
"val_bpb": 0.32778816957977247,
|
| 5 |
+
"total_batch_size": 2097152,
|
| 6 |
+
"model_config": {
|
| 7 |
+
"sequence_len": 4096,
|
| 8 |
+
"vocab_size": 32768,
|
| 9 |
+
"n_layer": 32,
|
| 10 |
+
"n_head": 16,
|
| 11 |
+
"n_kv_head": 16,
|
| 12 |
+
"n_embd": 2048,
|
| 13 |
+
"window_pattern": "SSSL"
|
| 14 |
+
},
|
| 15 |
+
"user_config": {
|
| 16 |
+
"run": "Think.Unbounded-d32-v2mix-cont-nanochat-default-datamatch-v1",
|
| 17 |
+
"wandb_run_id": "8c553682",
|
| 18 |
+
"wandb_group": "think-d32",
|
| 19 |
+
"wandb_tags": "sft,modern,nanochat-default,data-matched,compute-controlled,d32",
|
| 20 |
+
"device_type": "",
|
| 21 |
+
"model_tag": null,
|
| 22 |
+
"model_step": null,
|
| 23 |
+
"base_checkpoint_dir": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/base_checkpoints",
|
| 24 |
+
"base_step": 9600,
|
| 25 |
+
"checkpoint_dir": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-nanochat-default-datamatch-v1/checkpoints",
|
| 26 |
+
"tokenizer_dir": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/tokenizer",
|
| 27 |
+
"resume_from_step": null,
|
| 28 |
+
"experiment_id": "Think.Unbounded-d32-v2mix-cont-nanochat-default-datamatch-v1",
|
| 29 |
+
"experiment_config": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-nanochat-default-datamatch-v1/config.json",
|
| 30 |
+
"parent_cumulative_flops": 2.3103503196928082e+20,
|
| 31 |
+
"tokenizer_fingerprint": "21e99d5cdeeaa660",
|
| 32 |
+
"git_commit_sha": "3c34775d260a016a79ab46ea76f8c8e48458fd68",
|
| 33 |
+
"load_optimizer": 0,
|
| 34 |
+
"num_iterations": -1,
|
| 35 |
+
"num_epochs": 1,
|
| 36 |
+
"target_examples_per_step": 0,
|
| 37 |
+
"max_seq_len": null,
|
| 38 |
+
"device_batch_size": 2,
|
| 39 |
+
"total_batch_size": null,
|
| 40 |
+
"embedding_lr": null,
|
| 41 |
+
"unembedding_lr": null,
|
| 42 |
+
"matrix_lr": null,
|
| 43 |
+
"init_lr_frac": 0.8,
|
| 44 |
+
"warmup_ratio": 0.03,
|
| 45 |
+
"warmdown_ratio": 0.5,
|
| 46 |
+
"final_lr_frac": 0.0,
|
| 47 |
+
"eval_every": -1,
|
| 48 |
+
"eval_tokens": 20971520,
|
| 49 |
+
"chatcore_every": -1,
|
| 50 |
+
"chatcore_max_cat": -1,
|
| 51 |
+
"chatcore_max_sample": 32,
|
| 52 |
+
"save_every": -1,
|
| 53 |
+
"recipe": "nanochat-default",
|
| 54 |
+
"curriculum_config": "",
|
| 55 |
+
"pre1930_epochs": 5,
|
| 56 |
+
"mmlu_epochs": 3,
|
| 57 |
+
"gsm8k_epochs": 4,
|
| 58 |
+
"max_train_presentations": 652950,
|
| 59 |
+
"authentic_epochs": 0,
|
| 60 |
+
"knowledge_qa_epochs": 0,
|
| 61 |
+
"multiturn_qa_epochs": 0,
|
| 62 |
+
"reasoning_qa_epochs": 0,
|
| 63 |
+
"stem_reasoning_epochs": 0,
|
| 64 |
+
"narrative_grounded_epochs": 0,
|
| 65 |
+
"narrative_fiction_epochs": 0,
|
| 66 |
+
"opinion_qa_epochs": 0,
|
| 67 |
+
"how_to_qa_epochs": 0,
|
| 68 |
+
"verse_qa_epochs": 0,
|
| 69 |
+
"composition_qa_epochs": 0,
|
| 70 |
+
"resolved_experiment_config": {
|
| 71 |
+
"schema_version": 1,
|
| 72 |
+
"stage": "sft",
|
| 73 |
+
"experiment_suffix": "nanochat-default-datamatch-v1",
|
| 74 |
+
"data": {
|
| 75 |
+
"recipe": "nanochat-default",
|
| 76 |
+
"mmlu_epochs": 3,
|
| 77 |
+
"gsm8k_epochs": 4,
|
| 78 |
+
"max_train_presentations": 652950
|
| 79 |
+
},
|
| 80 |
+
"training": {
|
| 81 |
+
"num_iterations": -1,
|
| 82 |
+
"load_optimizer": 0,
|
| 83 |
+
"device_batch_size": 2,
|
| 84 |
+
"warmup_ratio": 0.03,
|
| 85 |
+
"warmdown_ratio": 0.5,
|
| 86 |
+
"init_lr_frac": 0.8,
|
| 87 |
+
"final_lr_frac": 0.0,
|
| 88 |
+
"eval_every": -1,
|
| 89 |
+
"chatcore_every": -1,
|
| 90 |
+
"save_every": -1
|
| 91 |
+
},
|
| 92 |
+
"comparison": {
|
| 93 |
+
"reference_sft": "pre1930-curriculum-c3-robust-v2",
|
| 94 |
+
"reference_train_presentations": 652950,
|
| 95 |
+
"reference_optimizer_steps": 42,
|
| 96 |
+
"reference_stage_training_flops": 1.0107782648656036e+18,
|
| 97 |
+
"matching_priority": "row_presentations",
|
| 98 |
+
"note": "Realized compute is measured because conversation lengths differ."
|
| 99 |
+
},
|
| 100 |
+
"artifacts": {
|
| 101 |
+
"repo": "jbduran/think.nano"
|
| 102 |
+
},
|
| 103 |
+
"wandb": {
|
| 104 |
+
"enabled": true,
|
| 105 |
+
"entity": "jbduran-thinkingmachinesncsu",
|
| 106 |
+
"project": "think.nano",
|
| 107 |
+
"group": "think-d32",
|
| 108 |
+
"tags": [
|
| 109 |
+
"sft",
|
| 110 |
+
"modern",
|
| 111 |
+
"nanochat-default",
|
| 112 |
+
"data-matched",
|
| 113 |
+
"compute-controlled",
|
| 114 |
+
"d32"
|
| 115 |
+
]
|
| 116 |
+
},
|
| 117 |
+
"config_fingerprint": "6359be7150ece075",
|
| 118 |
+
"artifact_path": "experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-nanochat-default-datamatch-v1"
|
| 119 |
+
},
|
| 120 |
+
"stage": "sft",
|
| 121 |
+
"base_experiment_id": null,
|
| 122 |
+
"parent_experiment_id": null,
|
| 123 |
+
"parent_checkpoint_step": null,
|
| 124 |
+
"config_fingerprint": "6359be7150ece075"
|
| 125 |
+
},
|
| 126 |
+
"sft_mixture_summary": {
|
| 127 |
+
"schema_version": 1,
|
| 128 |
+
"recipe": "nanochat-default",
|
| 129 |
+
"selection": {
|
| 130 |
+
"algorithm": "TaskMixture(seed=42) deterministic global shuffle then prefix",
|
| 131 |
+
"selected_presentations": 652950,
|
| 132 |
+
"full_mixture_presentations": 1071759,
|
| 133 |
+
"selection_sha256": "975f61958382acd3a3cd9b01c2f3204cff8982908365503ac8972e29d38bfec2"
|
| 134 |
+
},
|
| 135 |
+
"sources": {
|
| 136 |
+
"smoltalk": {
|
| 137 |
+
"kind": "huggingface",
|
| 138 |
+
"repo": "HuggingFaceTB/smol-smoltalk",
|
| 139 |
+
"split": "train",
|
| 140 |
+
"dataset_fingerprint": "7776c4bbcd5d614f",
|
| 141 |
+
"configured_repetitions": 1,
|
| 142 |
+
"available_rows": 460341,
|
| 143 |
+
"resolved_revision": "f73fe857d519ff6ac5af2ea67c4d3834da7b8bcc",
|
| 144 |
+
"full_mixture_presentations": 460341,
|
| 145 |
+
"selected_presentations": 280414,
|
| 146 |
+
"selected_presentations_by_replica": [
|
| 147 |
+
280414
|
| 148 |
+
],
|
| 149 |
+
"selected_distinct_rows": 280414,
|
| 150 |
+
"selected_fraction": 0.4294570794088368
|
| 151 |
+
},
|
| 152 |
+
"identity": {
|
| 153 |
+
"kind": "jsonl",
|
| 154 |
+
"url": "https://karpathy-public.s3.us-west-2.amazonaws.com/identity_conversations.jsonl",
|
| 155 |
+
"sha256": "85a6a96342b88addeb1b8267933c47a6e3ceb1dfc07b305b805fd7ff8106878e",
|
| 156 |
+
"configured_repetitions": 2,
|
| 157 |
+
"available_rows": 1000,
|
| 158 |
+
"full_mixture_presentations": 2000,
|
| 159 |
+
"selected_presentations": 1241,
|
| 160 |
+
"selected_presentations_by_replica": [
|
| 161 |
+
630,
|
| 162 |
+
611
|
| 163 |
+
],
|
| 164 |
+
"selected_distinct_rows": 874,
|
| 165 |
+
"selected_fraction": 0.0019006049467799986
|
| 166 |
+
},
|
| 167 |
+
"mmlu": {
|
| 168 |
+
"kind": "huggingface",
|
| 169 |
+
"repo": "cais/mmlu",
|
| 170 |
+
"subset": "all",
|
| 171 |
+
"split": "auxiliary_train",
|
| 172 |
+
"dataset_fingerprint": "5e98d840a8432dc6",
|
| 173 |
+
"configured_repetitions": 3,
|
| 174 |
+
"available_rows": 99842,
|
| 175 |
+
"resolved_revision": "c30699e8356da336a370243923dbaf21066bb9fe",
|
| 176 |
+
"full_mixture_presentations": 299526,
|
| 177 |
+
"selected_presentations": 182570,
|
| 178 |
+
"selected_presentations_by_replica": [
|
| 179 |
+
60835,
|
| 180 |
+
60932,
|
| 181 |
+
60803
|
| 182 |
+
],
|
| 183 |
+
"selected_distinct_rows": 93775,
|
| 184 |
+
"selected_fraction": 0.2796079332261276
|
| 185 |
+
},
|
| 186 |
+
"gsm8k": {
|
| 187 |
+
"kind": "huggingface",
|
| 188 |
+
"repo": "openai/gsm8k",
|
| 189 |
+
"subset": "main",
|
| 190 |
+
"split": "train",
|
| 191 |
+
"dataset_fingerprint": "8b93a690dcf7094e",
|
| 192 |
+
"configured_repetitions": 4,
|
| 193 |
+
"available_rows": 7473,
|
| 194 |
+
"resolved_revision": "740312add88f781978c0658806c59bc2815b9866",
|
| 195 |
+
"full_mixture_presentations": 29892,
|
| 196 |
+
"selected_presentations": 18182,
|
| 197 |
+
"selected_presentations_by_replica": [
|
| 198 |
+
4516,
|
| 199 |
+
4549,
|
| 200 |
+
4540,
|
| 201 |
+
4577
|
| 202 |
+
],
|
| 203 |
+
"selected_distinct_rows": 7288,
|
| 204 |
+
"selected_fraction": 0.02784593000995482
|
| 205 |
+
},
|
| 206 |
+
"simple_spelling": {
|
| 207 |
+
"kind": "deterministic_generated",
|
| 208 |
+
"generator": "tasks.spellingbee.SimpleSpelling",
|
| 209 |
+
"word_list_url": "https://raw.githubusercontent.com/dwyl/english-words/refs/heads/master/words_alpha.txt",
|
| 210 |
+
"word_list_sha256": "643c7f71d5caee9806be56f5ed83ba36e2246a06a04bcf4aaa0576790bb90d66",
|
| 211 |
+
"configured_repetitions": 1,
|
| 212 |
+
"available_rows": 200000,
|
| 213 |
+
"full_mixture_presentations": 200000,
|
| 214 |
+
"selected_presentations": 121661,
|
| 215 |
+
"selected_presentations_by_replica": [
|
| 216 |
+
121661
|
| 217 |
+
],
|
| 218 |
+
"selected_distinct_rows": 121661,
|
| 219 |
+
"selected_fraction": 0.18632513975036374
|
| 220 |
+
},
|
| 221 |
+
"spelling_bee": {
|
| 222 |
+
"kind": "deterministic_generated",
|
| 223 |
+
"generator": "tasks.spellingbee.SpellingBee",
|
| 224 |
+
"word_list_url": "https://raw.githubusercontent.com/dwyl/english-words/refs/heads/master/words_alpha.txt",
|
| 225 |
+
"word_list_sha256": "643c7f71d5caee9806be56f5ed83ba36e2246a06a04bcf4aaa0576790bb90d66",
|
| 226 |
+
"configured_repetitions": 1,
|
| 227 |
+
"available_rows": 80000,
|
| 228 |
+
"full_mixture_presentations": 80000,
|
| 229 |
+
"selected_presentations": 48882,
|
| 230 |
+
"selected_presentations_by_replica": [
|
| 231 |
+
48882
|
| 232 |
+
],
|
| 233 |
+
"selected_distinct_rows": 48882,
|
| 234 |
+
"selected_fraction": 0.07486331265793705
|
| 235 |
+
}
|
| 236 |
+
}
|
| 237 |
+
},
|
| 238 |
+
"loop_state": {
|
| 239 |
+
"step": 160,
|
| 240 |
+
"total_training_time": 6951.126858472824,
|
| 241 |
+
"min_val_bpb": 0.32778816957977247,
|
| 242 |
+
"smooth_train_loss": 0.8890764356615257,
|
| 243 |
+
"mfu": 52.42433466002523,
|
| 244 |
+
"tok_per_sec": 45180,
|
| 245 |
+
"stage_training_flops": 3.8505838661546803e+18,
|
| 246 |
+
"stage_training_tokens": 335544320,
|
| 247 |
+
"inherited_parent_flops": 2.3103503196928082e+20,
|
| 248 |
+
"cumulative_pipeline_training_flops": 2.348856158354355e+20
|
| 249 |
+
}
|
| 250 |
+
}
|