hosting: bf16 export for image bake (meta_000042.json)
Browse files
experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-pre1930-curriculum-c3-robust-v2/bf16/meta_000042.json
ADDED
|
@@ -0,0 +1,208 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"step": 42,
|
| 3 |
+
"training_complete": true,
|
| 4 |
+
"val_bpb": 0.7037352212342954,
|
| 5 |
+
"model_config": {
|
| 6 |
+
"sequence_len": 4096,
|
| 7 |
+
"vocab_size": 32768,
|
| 8 |
+
"n_layer": 32,
|
| 9 |
+
"n_head": 16,
|
| 10 |
+
"n_kv_head": 16,
|
| 11 |
+
"n_embd": 2048,
|
| 12 |
+
"window_pattern": "SSSL"
|
| 13 |
+
},
|
| 14 |
+
"user_config": {
|
| 15 |
+
"run": "Think.Unbounded-d32-v2mix-cont-pre1930-curriculum-c3-robust-v2",
|
| 16 |
+
"wandb_run_id": "036b1c71",
|
| 17 |
+
"wandb_group": "think-d32",
|
| 18 |
+
"wandb_tags": "sft,curriculum,c3,scale-max,staged,robustness,noise,d32",
|
| 19 |
+
"device_type": "",
|
| 20 |
+
"model_tag": null,
|
| 21 |
+
"model_step": null,
|
| 22 |
+
"base_checkpoint_dir": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/base_checkpoints",
|
| 23 |
+
"base_step": 9600,
|
| 24 |
+
"checkpoint_dir": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-pre1930-curriculum-c3-robust-v2/checkpoints",
|
| 25 |
+
"tokenizer_dir": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/tokenizer",
|
| 26 |
+
"resume_from_step": null,
|
| 27 |
+
"experiment_id": "Think.Unbounded-d32-v2mix-cont-pre1930-curriculum-c3-robust-v2",
|
| 28 |
+
"experiment_config": "/workspace/nanochat/experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-pre1930-curriculum-c3-robust-v2/config.json",
|
| 29 |
+
"parent_cumulative_flops": 2.3103503196928082e+20,
|
| 30 |
+
"tokenizer_fingerprint": "21e99d5cdeeaa660",
|
| 31 |
+
"git_commit_sha": "04bb043ad61d4db3e9022c422302df7b8f2dc0c9",
|
| 32 |
+
"load_optimizer": 0,
|
| 33 |
+
"num_iterations": -1,
|
| 34 |
+
"max_seq_len": null,
|
| 35 |
+
"device_batch_size": 2,
|
| 36 |
+
"total_batch_size": null,
|
| 37 |
+
"embedding_lr": null,
|
| 38 |
+
"unembedding_lr": null,
|
| 39 |
+
"matrix_lr": null,
|
| 40 |
+
"init_lr_frac": 0.8,
|
| 41 |
+
"warmup_ratio": 0.03,
|
| 42 |
+
"warmdown_ratio": 0.5,
|
| 43 |
+
"final_lr_frac": 0.0,
|
| 44 |
+
"eval_every": 400,
|
| 45 |
+
"eval_tokens": 20971520,
|
| 46 |
+
"chatcore_every": 100000,
|
| 47 |
+
"chatcore_max_cat": -1,
|
| 48 |
+
"chatcore_max_sample": 32,
|
| 49 |
+
"save_every": -1,
|
| 50 |
+
"recipe": "curriculum",
|
| 51 |
+
"curriculum_config": "",
|
| 52 |
+
"pre1930_epochs": 5,
|
| 53 |
+
"mmlu_epochs": 3,
|
| 54 |
+
"gsm8k_epochs": 4,
|
| 55 |
+
"authentic_epochs": 0,
|
| 56 |
+
"knowledge_qa_epochs": 0,
|
| 57 |
+
"multiturn_qa_epochs": 0,
|
| 58 |
+
"reasoning_qa_epochs": 0,
|
| 59 |
+
"stem_reasoning_epochs": 0,
|
| 60 |
+
"narrative_grounded_epochs": 0,
|
| 61 |
+
"narrative_fiction_epochs": 0,
|
| 62 |
+
"opinion_qa_epochs": 0,
|
| 63 |
+
"how_to_qa_epochs": 0,
|
| 64 |
+
"verse_qa_epochs": 0,
|
| 65 |
+
"composition_qa_epochs": 0,
|
| 66 |
+
"resolved_experiment_config": {
|
| 67 |
+
"schema_version": 1,
|
| 68 |
+
"stage": "sft",
|
| 69 |
+
"experiment_suffix": "pre1930-curriculum-c3-robust-v2",
|
| 70 |
+
"data": {
|
| 71 |
+
"recipe": "curriculum",
|
| 72 |
+
"curriculum": {
|
| 73 |
+
"name": "C3Rv2",
|
| 74 |
+
"mode": "staged",
|
| 75 |
+
"threshold_default": 80,
|
| 76 |
+
"stages": [
|
| 77 |
+
{
|
| 78 |
+
"routes": [
|
| 79 |
+
"knowledge_qa"
|
| 80 |
+
],
|
| 81 |
+
"authentic": "single"
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"routes": [
|
| 85 |
+
"reasoning_qa",
|
| 86 |
+
"stem_reasoning",
|
| 87 |
+
"how_to_qa",
|
| 88 |
+
"opinion_qa",
|
| 89 |
+
"composition_qa",
|
| 90 |
+
"verse_qa"
|
| 91 |
+
],
|
| 92 |
+
"calibration_qa": true
|
| 93 |
+
},
|
| 94 |
+
{
|
| 95 |
+
"routes": [
|
| 96 |
+
"multiturn_qa",
|
| 97 |
+
"narrative_grounded",
|
| 98 |
+
"narrative_fiction"
|
| 99 |
+
],
|
| 100 |
+
"authentic": "multi"
|
| 101 |
+
}
|
| 102 |
+
],
|
| 103 |
+
"noise": {
|
| 104 |
+
"rate": 0.3
|
| 105 |
+
},
|
| 106 |
+
"robustness": {
|
| 107 |
+
"stage": 0,
|
| 108 |
+
"epochs": 1,
|
| 109 |
+
"routes": {
|
| 110 |
+
"conversation_qa": {
|
| 111 |
+
"count": null
|
| 112 |
+
},
|
| 113 |
+
"unparseable_qa": {
|
| 114 |
+
"count": null
|
| 115 |
+
},
|
| 116 |
+
"typo_qa": {
|
| 117 |
+
"count": null
|
| 118 |
+
},
|
| 119 |
+
"era_qa": {
|
| 120 |
+
"count": null
|
| 121 |
+
},
|
| 122 |
+
"conversation_multiturn": {
|
| 123 |
+
"count": null
|
| 124 |
+
}
|
| 125 |
+
}
|
| 126 |
+
}
|
| 127 |
+
}
|
| 128 |
+
},
|
| 129 |
+
"training": {
|
| 130 |
+
"num_iterations": -1,
|
| 131 |
+
"load_optimizer": 0,
|
| 132 |
+
"device_batch_size": 2,
|
| 133 |
+
"warmup_ratio": 0.03,
|
| 134 |
+
"eval_every": 400,
|
| 135 |
+
"chatcore_every": 100000,
|
| 136 |
+
"chatcore_max_sample": 32,
|
| 137 |
+
"save_every": -1
|
| 138 |
+
},
|
| 139 |
+
"artifacts": {
|
| 140 |
+
"repo": "jbduran/think.nano"
|
| 141 |
+
},
|
| 142 |
+
"wandb": {
|
| 143 |
+
"enabled": true,
|
| 144 |
+
"entity": "jbduran-thinkingmachinesncsu",
|
| 145 |
+
"project": "think.nano",
|
| 146 |
+
"group": "think-d32",
|
| 147 |
+
"tags": [
|
| 148 |
+
"sft",
|
| 149 |
+
"curriculum",
|
| 150 |
+
"c3",
|
| 151 |
+
"scale-max",
|
| 152 |
+
"staged",
|
| 153 |
+
"robustness",
|
| 154 |
+
"noise",
|
| 155 |
+
"d32"
|
| 156 |
+
]
|
| 157 |
+
},
|
| 158 |
+
"config_fingerprint": "4462a389dc98cd5d",
|
| 159 |
+
"artifact_path": "experiments/Think.Unbounded-d32-v2mix-cont/sft/Think.Unbounded-d32-v2mix-cont-pre1930-curriculum-c3-robust-v2"
|
| 160 |
+
},
|
| 161 |
+
"stage": "sft",
|
| 162 |
+
"base_experiment_id": null,
|
| 163 |
+
"parent_experiment_id": null,
|
| 164 |
+
"parent_checkpoint_step": null,
|
| 165 |
+
"config_fingerprint": "4462a389dc98cd5d"
|
| 166 |
+
},
|
| 167 |
+
"loop_state": {
|
| 168 |
+
"step": 42,
|
| 169 |
+
"total_training_time": 1503.5729036331177,
|
| 170 |
+
"min_val_bpb": 0.7037352212342954,
|
| 171 |
+
"smooth_train_loss": 1.8593226867746961,
|
| 172 |
+
"mfu": 51.82901425419851,
|
| 173 |
+
"tok_per_sec": 44667,
|
| 174 |
+
"stage_training_flops": 1.0107782648656036e+18,
|
| 175 |
+
"inherited_parent_flops": 2.3103503196928082e+20,
|
| 176 |
+
"cumulative_pipeline_training_flops": 2.3204581023414642e+20
|
| 177 |
+
},
|
| 178 |
+
"export": {
|
| 179 |
+
"source_dir": "/content/ckpt",
|
| 180 |
+
"source_step": 42,
|
| 181 |
+
"storage_dtype": "bfloat16",
|
| 182 |
+
"cast_tensors": 210,
|
| 183 |
+
"kept_fp32_tensors": [
|
| 184 |
+
"resid_lambdas",
|
| 185 |
+
"x0_lambdas",
|
| 186 |
+
"smear_lambda",
|
| 187 |
+
"backout_lambda",
|
| 188 |
+
"transformer.wte.weight",
|
| 189 |
+
"value_embeds.1.weight",
|
| 190 |
+
"value_embeds.3.weight",
|
| 191 |
+
"value_embeds.5.weight",
|
| 192 |
+
"value_embeds.7.weight",
|
| 193 |
+
"value_embeds.9.weight",
|
| 194 |
+
"value_embeds.11.weight",
|
| 195 |
+
"value_embeds.13.weight",
|
| 196 |
+
"value_embeds.15.weight",
|
| 197 |
+
"value_embeds.17.weight",
|
| 198 |
+
"value_embeds.19.weight",
|
| 199 |
+
"value_embeds.21.weight",
|
| 200 |
+
"value_embeds.23.weight",
|
| 201 |
+
"value_embeds.25.weight",
|
| 202 |
+
"value_embeds.27.weight",
|
| 203 |
+
"value_embeds.29.weight",
|
| 204 |
+
"value_embeds.31.weight"
|
| 205 |
+
],
|
| 206 |
+
"optimizer_state_stripped": true
|
| 207 |
+
}
|
| 208 |
+
}
|