Auto upload 2026-08-09T07:04:11.258775 (part 2)
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- out/glu-waleed-94L_run/checkpoint-600/rng_state.pth +3 -0
- out/glu-waleed-94L_run/checkpoint-600/scheduler.pt +3 -0
- out/glu-waleed-94L_run/checkpoint-600/tokenizer.json +0 -0
- out/glu-waleed-94L_run/checkpoint-600/tokenizer_config.json +13 -0
- out/glu-waleed-94L_run/checkpoint-600/trainer_state.json +340 -0
- out/glu-waleed-94L_run/checkpoint-600/training_args.bin +3 -0
- out/glu-waleed-94L_run/checkpoint-700/config.json +35 -0
- out/glu-waleed-94L_run/checkpoint-700/model.safetensors +3 -0
- out/glu-waleed-94L_run/checkpoint-700/optimizer.pt +3 -0
- out/glu-waleed-94L_run/checkpoint-700/rng_state.pth +3 -0
- out/glu-waleed-94L_run/checkpoint-700/scheduler.pt +3 -0
- out/glu-waleed-94L_run/checkpoint-700/tokenizer.json +0 -0
- out/glu-waleed-94L_run/checkpoint-700/tokenizer_config.json +13 -0
- out/glu-waleed-94L_run/checkpoint-700/trainer_state.json +391 -0
- out/glu-waleed-94L_run/checkpoint-700/training_args.bin +3 -0
- out/glu-waleed-94L_run/checkpoint-800/config.json +35 -0
- out/glu-waleed-94L_run/checkpoint-800/model.safetensors +3 -0
- out/glu-waleed-94L_run/checkpoint-800/optimizer.pt +3 -0
- out/glu-waleed-94L_run/checkpoint-800/rng_state.pth +3 -0
- out/glu-waleed-94L_run/checkpoint-800/scheduler.pt +3 -0
- out/glu-waleed-94L_run/checkpoint-800/tokenizer.json +0 -0
- out/glu-waleed-94L_run/checkpoint-800/tokenizer_config.json +13 -0
- out/glu-waleed-94L_run/checkpoint-800/trainer_state.json +442 -0
- out/glu-waleed-94L_run/checkpoint-800/training_args.bin +3 -0
- out/glu-waleed-94L_run/checkpoint-900/config.json +35 -0
- out/glu-waleed-94L_run/checkpoint-900/model.safetensors +3 -0
- out/glu-waleed-94L_run/checkpoint-900/optimizer.pt +3 -0
- out/glu-waleed-94L_run/checkpoint-900/rng_state.pth +3 -0
- out/glu-waleed-94L_run/checkpoint-900/scheduler.pt +3 -0
- out/glu-waleed-94L_run/checkpoint-900/tokenizer.json +0 -0
- out/glu-waleed-94L_run/checkpoint-900/tokenizer_config.json +13 -0
- out/glu-waleed-94L_run/checkpoint-900/trainer_state.json +493 -0
- out/glu-waleed-94L_run/checkpoint-900/training_args.bin +3 -0
- out/glu-waleed-94L_run/config.json +35 -0
- out/glu-waleed-94L_run/model.safetensors +3 -0
- out/glu-waleed-94L_run/tokenizer.json +0 -0
- out/glu-waleed-94L_run/tokenizer_config.json +13 -0
- out/glu-waleed-94L_run/training_args.bin +3 -0
- out/glu-waleed-94L_run/training_log.jsonl +0 -0
- out/mlp-s10-94L_run/README.md +87 -0
- out/mlp-s10-94L_run/checkpoint-100/config.json +35 -0
- out/mlp-s10-94L_run/checkpoint-100/model.safetensors +3 -0
- out/mlp-s10-94L_run/checkpoint-100/optimizer.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-100/rng_state.pth +3 -0
- out/mlp-s10-94L_run/checkpoint-100/scheduler.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-100/tokenizer.json +0 -0
- out/mlp-s10-94L_run/checkpoint-100/tokenizer_config.json +13 -0
- out/mlp-s10-94L_run/checkpoint-100/trainer_state.json +85 -0
- out/mlp-s10-94L_run/checkpoint-100/training_args.bin +3 -0
- out/mlp-s10-94L_run/checkpoint-1000/config.json +35 -0
out/glu-waleed-94L_run/checkpoint-600/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b0b8a6e3878f9402461193b241d8f1ee8546c6ad03f43e5b9dab8f4fc8c4d065
|
| 3 |
+
size 14244
|
out/glu-waleed-94L_run/checkpoint-600/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c5502244c7ba8b61ab0a3b01cdc7e5e2629060a0565d6b782445d6ba31022cee
|
| 3 |
+
size 1064
|
out/glu-waleed-94L_run/checkpoint-600/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/glu-waleed-94L_run/checkpoint-600/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/glu-waleed-94L_run/checkpoint-600/trainer_state.json
ADDED
|
@@ -0,0 +1,340 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.3235373416015098,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 600,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.625,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.4543212890625,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.2314453125,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.255674743652342,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.911471843719482,
|
| 29 |
+
"eval_runtime": 20.9396,
|
| 30 |
+
"eval_samples_per_second": 454.975,
|
| 31 |
+
"eval_steps_per_second": 3.582,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 8.75,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.663941955566408,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 4.15625,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 23.1123046875,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 1.578125,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.97301483154297,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.725912094116211,
|
| 58 |
+
"eval_runtime": 20.3952,
|
| 59 |
+
"eval_samples_per_second": 467.12,
|
| 60 |
+
"eval_steps_per_second": 3.677,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 0.2099609375,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 22.912661743164062,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 0.859375,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 22.89833984375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.71552038192749,
|
| 80 |
+
"eval_runtime": 20.5186,
|
| 81 |
+
"eval_samples_per_second": 464.31,
|
| 82 |
+
"eval_steps_per_second": 3.655,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 5.3125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 22.871157836914062,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 3.515625,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 22.64031982421875,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 2.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 21.882586669921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 5.379149913787842,
|
| 109 |
+
"eval_runtime": 19.954,
|
| 110 |
+
"eval_samples_per_second": 477.447,
|
| 111 |
+
"eval_steps_per_second": 3.759,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.11863035858722028,
|
| 116 |
+
"grad_norm": 3.0,
|
| 117 |
+
"learning_rate": 0.001,
|
| 118 |
+
"loss": 21.073570251464844,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.12941493664060394,
|
| 123 |
+
"grad_norm": 26.625,
|
| 124 |
+
"learning_rate": 0.001,
|
| 125 |
+
"loss": 20.38078918457031,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.13480722566729578,
|
| 130 |
+
"eval_loss": 4.969682693481445,
|
| 131 |
+
"eval_runtime": 20.8528,
|
| 132 |
+
"eval_samples_per_second": 456.87,
|
| 133 |
+
"eval_steps_per_second": 3.597,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.1401995146939876,
|
| 138 |
+
"grad_norm": 5.5625,
|
| 139 |
+
"learning_rate": 0.001,
|
| 140 |
+
"loss": 19.993435668945313,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.15098409274737126,
|
| 145 |
+
"grad_norm": 3.84375,
|
| 146 |
+
"learning_rate": 0.001,
|
| 147 |
+
"loss": 19.312832641601563,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.1617686708007549,
|
| 152 |
+
"grad_norm": 8.1875,
|
| 153 |
+
"learning_rate": 0.001,
|
| 154 |
+
"loss": 18.880908203125,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.1617686708007549,
|
| 159 |
+
"eval_loss": 4.66710090637207,
|
| 160 |
+
"eval_runtime": 20.0053,
|
| 161 |
+
"eval_samples_per_second": 476.224,
|
| 162 |
+
"eval_steps_per_second": 3.749,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.17255324885413859,
|
| 167 |
+
"grad_norm": 8.6875,
|
| 168 |
+
"learning_rate": 0.001,
|
| 169 |
+
"loss": 18.437432861328126,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.18333782690752223,
|
| 174 |
+
"grad_norm": 8.0,
|
| 175 |
+
"learning_rate": 0.001,
|
| 176 |
+
"loss": 18.02788543701172,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.18873011593421407,
|
| 181 |
+
"eval_loss": 4.424458980560303,
|
| 182 |
+
"eval_runtime": 20.6161,
|
| 183 |
+
"eval_samples_per_second": 462.114,
|
| 184 |
+
"eval_steps_per_second": 3.638,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.1941224049609059,
|
| 189 |
+
"grad_norm": 9.9375,
|
| 190 |
+
"learning_rate": 0.001,
|
| 191 |
+
"loss": 17.731655883789063,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.20490698301428956,
|
| 196 |
+
"grad_norm": 15.625,
|
| 197 |
+
"learning_rate": 0.001,
|
| 198 |
+
"loss": 17.360447692871094,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.21569156106767323,
|
| 203 |
+
"grad_norm": 4.71875,
|
| 204 |
+
"learning_rate": 0.001,
|
| 205 |
+
"loss": 17.05499267578125,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.21569156106767323,
|
| 210 |
+
"eval_loss": 4.228377342224121,
|
| 211 |
+
"eval_runtime": 20.6518,
|
| 212 |
+
"eval_samples_per_second": 461.316,
|
| 213 |
+
"eval_steps_per_second": 3.632,
|
| 214 |
+
"step": 400
|
| 215 |
+
},
|
| 216 |
+
{
|
| 217 |
+
"epoch": 0.22647613912105688,
|
| 218 |
+
"grad_norm": 3.03125,
|
| 219 |
+
"learning_rate": 0.001,
|
| 220 |
+
"loss": 16.765740966796876,
|
| 221 |
+
"step": 420
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 0.23726071717444056,
|
| 225 |
+
"grad_norm": 16.625,
|
| 226 |
+
"learning_rate": 0.001,
|
| 227 |
+
"loss": 16.70984344482422,
|
| 228 |
+
"step": 440
|
| 229 |
+
},
|
| 230 |
+
{
|
| 231 |
+
"epoch": 0.2426530062011324,
|
| 232 |
+
"eval_loss": 4.115875244140625,
|
| 233 |
+
"eval_runtime": 17.5037,
|
| 234 |
+
"eval_samples_per_second": 544.283,
|
| 235 |
+
"eval_steps_per_second": 4.285,
|
| 236 |
+
"step": 450
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.2480452952278242,
|
| 240 |
+
"grad_norm": 9.625,
|
| 241 |
+
"learning_rate": 0.001,
|
| 242 |
+
"loss": 16.492950439453125,
|
| 243 |
+
"step": 460
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"epoch": 0.2588298732812079,
|
| 247 |
+
"grad_norm": 4.4375,
|
| 248 |
+
"learning_rate": 0.001,
|
| 249 |
+
"loss": 16.204039001464842,
|
| 250 |
+
"step": 480
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.26961445133459155,
|
| 254 |
+
"grad_norm": 3.21875,
|
| 255 |
+
"learning_rate": 0.001,
|
| 256 |
+
"loss": 15.97710723876953,
|
| 257 |
+
"step": 500
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"epoch": 0.26961445133459155,
|
| 261 |
+
"eval_loss": 3.9576539993286133,
|
| 262 |
+
"eval_runtime": 16.9826,
|
| 263 |
+
"eval_samples_per_second": 560.986,
|
| 264 |
+
"eval_steps_per_second": 4.416,
|
| 265 |
+
"step": 500
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.2803990293879752,
|
| 269 |
+
"grad_norm": 10.5,
|
| 270 |
+
"learning_rate": 0.001,
|
| 271 |
+
"loss": 15.733917236328125,
|
| 272 |
+
"step": 520
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.29118360744135885,
|
| 276 |
+
"grad_norm": 9.8125,
|
| 277 |
+
"learning_rate": 0.001,
|
| 278 |
+
"loss": 15.579618835449219,
|
| 279 |
+
"step": 540
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.2965758964680507,
|
| 283 |
+
"eval_loss": 3.861560821533203,
|
| 284 |
+
"eval_runtime": 16.9704,
|
| 285 |
+
"eval_samples_per_second": 561.389,
|
| 286 |
+
"eval_steps_per_second": 4.419,
|
| 287 |
+
"step": 550
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"epoch": 0.3019681854947425,
|
| 291 |
+
"grad_norm": 3.984375,
|
| 292 |
+
"learning_rate": 0.001,
|
| 293 |
+
"loss": 15.382864379882813,
|
| 294 |
+
"step": 560
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.3127527635481262,
|
| 298 |
+
"grad_norm": 2.75,
|
| 299 |
+
"learning_rate": 0.001,
|
| 300 |
+
"loss": 15.098277282714843,
|
| 301 |
+
"step": 580
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"epoch": 0.3235373416015098,
|
| 305 |
+
"grad_norm": 5.75,
|
| 306 |
+
"learning_rate": 0.001,
|
| 307 |
+
"loss": 14.87418212890625,
|
| 308 |
+
"step": 600
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"epoch": 0.3235373416015098,
|
| 312 |
+
"eval_loss": 3.683532238006592,
|
| 313 |
+
"eval_runtime": 17.1359,
|
| 314 |
+
"eval_samples_per_second": 555.968,
|
| 315 |
+
"eval_steps_per_second": 4.377,
|
| 316 |
+
"step": 600
|
| 317 |
+
}
|
| 318 |
+
],
|
| 319 |
+
"logging_steps": 20,
|
| 320 |
+
"max_steps": 1500,
|
| 321 |
+
"num_input_tokens_seen": 0,
|
| 322 |
+
"num_train_epochs": 1,
|
| 323 |
+
"save_steps": 100,
|
| 324 |
+
"stateful_callbacks": {
|
| 325 |
+
"TrainerControl": {
|
| 326 |
+
"args": {
|
| 327 |
+
"should_epoch_stop": false,
|
| 328 |
+
"should_evaluate": false,
|
| 329 |
+
"should_log": false,
|
| 330 |
+
"should_save": true,
|
| 331 |
+
"should_training_stop": false
|
| 332 |
+
},
|
| 333 |
+
"attributes": {}
|
| 334 |
+
}
|
| 335 |
+
},
|
| 336 |
+
"total_flos": 1.45569997651968e+16,
|
| 337 |
+
"train_batch_size": 128,
|
| 338 |
+
"trial_name": null,
|
| 339 |
+
"trial_params": null
|
| 340 |
+
}
|
out/glu-waleed-94L_run/checkpoint-600/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e12a35f98937379bc279e337a2141ffbc223c6273ca4783d094a415f227e198
|
| 3 |
+
size 4920
|
out/glu-waleed-94L_run/checkpoint-700/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "waleed",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/glu-waleed-94L_run/checkpoint-700/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1785022fb6baa8b8074d19434738583cb426f88dfd1e61c2a5d7faafc6494452
|
| 3 |
+
size 31992520
|
out/glu-waleed-94L_run/checkpoint-700/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cbc94a350a24d9fdb3e00a8750372ee155b281ba3285de5269b50be144718b16
|
| 3 |
+
size 64520514
|
out/glu-waleed-94L_run/checkpoint-700/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c07c9483d2aaa0e0aa4859fa05bacc55a60e0f30ff9c95a3b76854e880483a96
|
| 3 |
+
size 14244
|
out/glu-waleed-94L_run/checkpoint-700/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3173cd6908322b02fdacbfeb79bbb6a1a1a40d218262cceb79ee8c11c4fa07d7
|
| 3 |
+
size 1064
|
out/glu-waleed-94L_run/checkpoint-700/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/glu-waleed-94L_run/checkpoint-700/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/glu-waleed-94L_run/checkpoint-700/trainer_state.json
ADDED
|
@@ -0,0 +1,391 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.37746023186842814,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 700,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.625,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.4543212890625,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.2314453125,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.255674743652342,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.911471843719482,
|
| 29 |
+
"eval_runtime": 20.9396,
|
| 30 |
+
"eval_samples_per_second": 454.975,
|
| 31 |
+
"eval_steps_per_second": 3.582,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 8.75,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.663941955566408,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 4.15625,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 23.1123046875,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 1.578125,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.97301483154297,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.725912094116211,
|
| 58 |
+
"eval_runtime": 20.3952,
|
| 59 |
+
"eval_samples_per_second": 467.12,
|
| 60 |
+
"eval_steps_per_second": 3.677,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 0.2099609375,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 22.912661743164062,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 0.859375,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 22.89833984375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.71552038192749,
|
| 80 |
+
"eval_runtime": 20.5186,
|
| 81 |
+
"eval_samples_per_second": 464.31,
|
| 82 |
+
"eval_steps_per_second": 3.655,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 5.3125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 22.871157836914062,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 3.515625,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 22.64031982421875,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 2.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 21.882586669921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 5.379149913787842,
|
| 109 |
+
"eval_runtime": 19.954,
|
| 110 |
+
"eval_samples_per_second": 477.447,
|
| 111 |
+
"eval_steps_per_second": 3.759,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.11863035858722028,
|
| 116 |
+
"grad_norm": 3.0,
|
| 117 |
+
"learning_rate": 0.001,
|
| 118 |
+
"loss": 21.073570251464844,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.12941493664060394,
|
| 123 |
+
"grad_norm": 26.625,
|
| 124 |
+
"learning_rate": 0.001,
|
| 125 |
+
"loss": 20.38078918457031,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.13480722566729578,
|
| 130 |
+
"eval_loss": 4.969682693481445,
|
| 131 |
+
"eval_runtime": 20.8528,
|
| 132 |
+
"eval_samples_per_second": 456.87,
|
| 133 |
+
"eval_steps_per_second": 3.597,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.1401995146939876,
|
| 138 |
+
"grad_norm": 5.5625,
|
| 139 |
+
"learning_rate": 0.001,
|
| 140 |
+
"loss": 19.993435668945313,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.15098409274737126,
|
| 145 |
+
"grad_norm": 3.84375,
|
| 146 |
+
"learning_rate": 0.001,
|
| 147 |
+
"loss": 19.312832641601563,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.1617686708007549,
|
| 152 |
+
"grad_norm": 8.1875,
|
| 153 |
+
"learning_rate": 0.001,
|
| 154 |
+
"loss": 18.880908203125,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.1617686708007549,
|
| 159 |
+
"eval_loss": 4.66710090637207,
|
| 160 |
+
"eval_runtime": 20.0053,
|
| 161 |
+
"eval_samples_per_second": 476.224,
|
| 162 |
+
"eval_steps_per_second": 3.749,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.17255324885413859,
|
| 167 |
+
"grad_norm": 8.6875,
|
| 168 |
+
"learning_rate": 0.001,
|
| 169 |
+
"loss": 18.437432861328126,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.18333782690752223,
|
| 174 |
+
"grad_norm": 8.0,
|
| 175 |
+
"learning_rate": 0.001,
|
| 176 |
+
"loss": 18.02788543701172,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.18873011593421407,
|
| 181 |
+
"eval_loss": 4.424458980560303,
|
| 182 |
+
"eval_runtime": 20.6161,
|
| 183 |
+
"eval_samples_per_second": 462.114,
|
| 184 |
+
"eval_steps_per_second": 3.638,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.1941224049609059,
|
| 189 |
+
"grad_norm": 9.9375,
|
| 190 |
+
"learning_rate": 0.001,
|
| 191 |
+
"loss": 17.731655883789063,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.20490698301428956,
|
| 196 |
+
"grad_norm": 15.625,
|
| 197 |
+
"learning_rate": 0.001,
|
| 198 |
+
"loss": 17.360447692871094,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.21569156106767323,
|
| 203 |
+
"grad_norm": 4.71875,
|
| 204 |
+
"learning_rate": 0.001,
|
| 205 |
+
"loss": 17.05499267578125,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.21569156106767323,
|
| 210 |
+
"eval_loss": 4.228377342224121,
|
| 211 |
+
"eval_runtime": 20.6518,
|
| 212 |
+
"eval_samples_per_second": 461.316,
|
| 213 |
+
"eval_steps_per_second": 3.632,
|
| 214 |
+
"step": 400
|
| 215 |
+
},
|
| 216 |
+
{
|
| 217 |
+
"epoch": 0.22647613912105688,
|
| 218 |
+
"grad_norm": 3.03125,
|
| 219 |
+
"learning_rate": 0.001,
|
| 220 |
+
"loss": 16.765740966796876,
|
| 221 |
+
"step": 420
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 0.23726071717444056,
|
| 225 |
+
"grad_norm": 16.625,
|
| 226 |
+
"learning_rate": 0.001,
|
| 227 |
+
"loss": 16.70984344482422,
|
| 228 |
+
"step": 440
|
| 229 |
+
},
|
| 230 |
+
{
|
| 231 |
+
"epoch": 0.2426530062011324,
|
| 232 |
+
"eval_loss": 4.115875244140625,
|
| 233 |
+
"eval_runtime": 17.5037,
|
| 234 |
+
"eval_samples_per_second": 544.283,
|
| 235 |
+
"eval_steps_per_second": 4.285,
|
| 236 |
+
"step": 450
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.2480452952278242,
|
| 240 |
+
"grad_norm": 9.625,
|
| 241 |
+
"learning_rate": 0.001,
|
| 242 |
+
"loss": 16.492950439453125,
|
| 243 |
+
"step": 460
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"epoch": 0.2588298732812079,
|
| 247 |
+
"grad_norm": 4.4375,
|
| 248 |
+
"learning_rate": 0.001,
|
| 249 |
+
"loss": 16.204039001464842,
|
| 250 |
+
"step": 480
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.26961445133459155,
|
| 254 |
+
"grad_norm": 3.21875,
|
| 255 |
+
"learning_rate": 0.001,
|
| 256 |
+
"loss": 15.97710723876953,
|
| 257 |
+
"step": 500
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"epoch": 0.26961445133459155,
|
| 261 |
+
"eval_loss": 3.9576539993286133,
|
| 262 |
+
"eval_runtime": 16.9826,
|
| 263 |
+
"eval_samples_per_second": 560.986,
|
| 264 |
+
"eval_steps_per_second": 4.416,
|
| 265 |
+
"step": 500
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.2803990293879752,
|
| 269 |
+
"grad_norm": 10.5,
|
| 270 |
+
"learning_rate": 0.001,
|
| 271 |
+
"loss": 15.733917236328125,
|
| 272 |
+
"step": 520
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.29118360744135885,
|
| 276 |
+
"grad_norm": 9.8125,
|
| 277 |
+
"learning_rate": 0.001,
|
| 278 |
+
"loss": 15.579618835449219,
|
| 279 |
+
"step": 540
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.2965758964680507,
|
| 283 |
+
"eval_loss": 3.861560821533203,
|
| 284 |
+
"eval_runtime": 16.9704,
|
| 285 |
+
"eval_samples_per_second": 561.389,
|
| 286 |
+
"eval_steps_per_second": 4.419,
|
| 287 |
+
"step": 550
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"epoch": 0.3019681854947425,
|
| 291 |
+
"grad_norm": 3.984375,
|
| 292 |
+
"learning_rate": 0.001,
|
| 293 |
+
"loss": 15.382864379882813,
|
| 294 |
+
"step": 560
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.3127527635481262,
|
| 298 |
+
"grad_norm": 2.75,
|
| 299 |
+
"learning_rate": 0.001,
|
| 300 |
+
"loss": 15.098277282714843,
|
| 301 |
+
"step": 580
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"epoch": 0.3235373416015098,
|
| 305 |
+
"grad_norm": 5.75,
|
| 306 |
+
"learning_rate": 0.001,
|
| 307 |
+
"loss": 14.87418212890625,
|
| 308 |
+
"step": 600
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"epoch": 0.3235373416015098,
|
| 312 |
+
"eval_loss": 3.683532238006592,
|
| 313 |
+
"eval_runtime": 17.1359,
|
| 314 |
+
"eval_samples_per_second": 555.968,
|
| 315 |
+
"eval_steps_per_second": 4.377,
|
| 316 |
+
"step": 600
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"epoch": 0.3343219196548935,
|
| 320 |
+
"grad_norm": 5.28125,
|
| 321 |
+
"learning_rate": 0.001,
|
| 322 |
+
"loss": 14.851046752929687,
|
| 323 |
+
"step": 620
|
| 324 |
+
},
|
| 325 |
+
{
|
| 326 |
+
"epoch": 0.34510649770827717,
|
| 327 |
+
"grad_norm": 5.125,
|
| 328 |
+
"learning_rate": 0.001,
|
| 329 |
+
"loss": 14.559681701660157,
|
| 330 |
+
"step": 640
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.350498786734969,
|
| 334 |
+
"eval_loss": 3.5849404335021973,
|
| 335 |
+
"eval_runtime": 16.9858,
|
| 336 |
+
"eval_samples_per_second": 560.88,
|
| 337 |
+
"eval_steps_per_second": 4.415,
|
| 338 |
+
"step": 650
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.35589107576166085,
|
| 342 |
+
"grad_norm": 7.375,
|
| 343 |
+
"learning_rate": 0.001,
|
| 344 |
+
"loss": 14.377513122558593,
|
| 345 |
+
"step": 660
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.36667565381504447,
|
| 349 |
+
"grad_norm": 4.0625,
|
| 350 |
+
"learning_rate": 0.001,
|
| 351 |
+
"loss": 14.217088317871093,
|
| 352 |
+
"step": 680
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.37746023186842814,
|
| 356 |
+
"grad_norm": 5.0625,
|
| 357 |
+
"learning_rate": 0.001,
|
| 358 |
+
"loss": 14.131674194335938,
|
| 359 |
+
"step": 700
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.37746023186842814,
|
| 363 |
+
"eval_loss": 3.5239312648773193,
|
| 364 |
+
"eval_runtime": 16.9182,
|
| 365 |
+
"eval_samples_per_second": 563.122,
|
| 366 |
+
"eval_steps_per_second": 4.433,
|
| 367 |
+
"step": 700
|
| 368 |
+
}
|
| 369 |
+
],
|
| 370 |
+
"logging_steps": 20,
|
| 371 |
+
"max_steps": 1500,
|
| 372 |
+
"num_input_tokens_seen": 0,
|
| 373 |
+
"num_train_epochs": 1,
|
| 374 |
+
"save_steps": 100,
|
| 375 |
+
"stateful_callbacks": {
|
| 376 |
+
"TrainerControl": {
|
| 377 |
+
"args": {
|
| 378 |
+
"should_epoch_stop": false,
|
| 379 |
+
"should_evaluate": false,
|
| 380 |
+
"should_log": false,
|
| 381 |
+
"should_save": true,
|
| 382 |
+
"should_training_stop": false
|
| 383 |
+
},
|
| 384 |
+
"attributes": {}
|
| 385 |
+
}
|
| 386 |
+
},
|
| 387 |
+
"total_flos": 1.69831663927296e+16,
|
| 388 |
+
"train_batch_size": 128,
|
| 389 |
+
"trial_name": null,
|
| 390 |
+
"trial_params": null
|
| 391 |
+
}
|
out/glu-waleed-94L_run/checkpoint-700/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e12a35f98937379bc279e337a2141ffbc223c6273ca4783d094a415f227e198
|
| 3 |
+
size 4920
|
out/glu-waleed-94L_run/checkpoint-800/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "waleed",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/glu-waleed-94L_run/checkpoint-800/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:46fe89c5895ce86178a81914fc825cac7ff8620147e8972fb61dcd35408614da
|
| 3 |
+
size 31992520
|
out/glu-waleed-94L_run/checkpoint-800/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d93022f222f8067d4d9d327b367b7fc9d2cdb9d6df2bc97c3395fce5df171b79
|
| 3 |
+
size 64520514
|
out/glu-waleed-94L_run/checkpoint-800/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f83fe17529e572dff2734bb21512b28dd7cf5d20ef0e84688f5068ffbf24e765
|
| 3 |
+
size 14244
|
out/glu-waleed-94L_run/checkpoint-800/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7334026bbd250befa466dcbe6bec8bfe84c03024263e4dea55e40698f2caa92e
|
| 3 |
+
size 1064
|
out/glu-waleed-94L_run/checkpoint-800/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/glu-waleed-94L_run/checkpoint-800/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/glu-waleed-94L_run/checkpoint-800/trainer_state.json
ADDED
|
@@ -0,0 +1,442 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.43138312213534646,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 800,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.625,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.4543212890625,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.2314453125,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.255674743652342,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.911471843719482,
|
| 29 |
+
"eval_runtime": 20.9396,
|
| 30 |
+
"eval_samples_per_second": 454.975,
|
| 31 |
+
"eval_steps_per_second": 3.582,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 8.75,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.663941955566408,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 4.15625,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 23.1123046875,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 1.578125,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.97301483154297,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.725912094116211,
|
| 58 |
+
"eval_runtime": 20.3952,
|
| 59 |
+
"eval_samples_per_second": 467.12,
|
| 60 |
+
"eval_steps_per_second": 3.677,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 0.2099609375,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 22.912661743164062,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 0.859375,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 22.89833984375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.71552038192749,
|
| 80 |
+
"eval_runtime": 20.5186,
|
| 81 |
+
"eval_samples_per_second": 464.31,
|
| 82 |
+
"eval_steps_per_second": 3.655,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 5.3125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 22.871157836914062,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 3.515625,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 22.64031982421875,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 2.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 21.882586669921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 5.379149913787842,
|
| 109 |
+
"eval_runtime": 19.954,
|
| 110 |
+
"eval_samples_per_second": 477.447,
|
| 111 |
+
"eval_steps_per_second": 3.759,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.11863035858722028,
|
| 116 |
+
"grad_norm": 3.0,
|
| 117 |
+
"learning_rate": 0.001,
|
| 118 |
+
"loss": 21.073570251464844,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.12941493664060394,
|
| 123 |
+
"grad_norm": 26.625,
|
| 124 |
+
"learning_rate": 0.001,
|
| 125 |
+
"loss": 20.38078918457031,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.13480722566729578,
|
| 130 |
+
"eval_loss": 4.969682693481445,
|
| 131 |
+
"eval_runtime": 20.8528,
|
| 132 |
+
"eval_samples_per_second": 456.87,
|
| 133 |
+
"eval_steps_per_second": 3.597,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.1401995146939876,
|
| 138 |
+
"grad_norm": 5.5625,
|
| 139 |
+
"learning_rate": 0.001,
|
| 140 |
+
"loss": 19.993435668945313,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.15098409274737126,
|
| 145 |
+
"grad_norm": 3.84375,
|
| 146 |
+
"learning_rate": 0.001,
|
| 147 |
+
"loss": 19.312832641601563,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.1617686708007549,
|
| 152 |
+
"grad_norm": 8.1875,
|
| 153 |
+
"learning_rate": 0.001,
|
| 154 |
+
"loss": 18.880908203125,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.1617686708007549,
|
| 159 |
+
"eval_loss": 4.66710090637207,
|
| 160 |
+
"eval_runtime": 20.0053,
|
| 161 |
+
"eval_samples_per_second": 476.224,
|
| 162 |
+
"eval_steps_per_second": 3.749,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.17255324885413859,
|
| 167 |
+
"grad_norm": 8.6875,
|
| 168 |
+
"learning_rate": 0.001,
|
| 169 |
+
"loss": 18.437432861328126,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.18333782690752223,
|
| 174 |
+
"grad_norm": 8.0,
|
| 175 |
+
"learning_rate": 0.001,
|
| 176 |
+
"loss": 18.02788543701172,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.18873011593421407,
|
| 181 |
+
"eval_loss": 4.424458980560303,
|
| 182 |
+
"eval_runtime": 20.6161,
|
| 183 |
+
"eval_samples_per_second": 462.114,
|
| 184 |
+
"eval_steps_per_second": 3.638,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.1941224049609059,
|
| 189 |
+
"grad_norm": 9.9375,
|
| 190 |
+
"learning_rate": 0.001,
|
| 191 |
+
"loss": 17.731655883789063,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.20490698301428956,
|
| 196 |
+
"grad_norm": 15.625,
|
| 197 |
+
"learning_rate": 0.001,
|
| 198 |
+
"loss": 17.360447692871094,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.21569156106767323,
|
| 203 |
+
"grad_norm": 4.71875,
|
| 204 |
+
"learning_rate": 0.001,
|
| 205 |
+
"loss": 17.05499267578125,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.21569156106767323,
|
| 210 |
+
"eval_loss": 4.228377342224121,
|
| 211 |
+
"eval_runtime": 20.6518,
|
| 212 |
+
"eval_samples_per_second": 461.316,
|
| 213 |
+
"eval_steps_per_second": 3.632,
|
| 214 |
+
"step": 400
|
| 215 |
+
},
|
| 216 |
+
{
|
| 217 |
+
"epoch": 0.22647613912105688,
|
| 218 |
+
"grad_norm": 3.03125,
|
| 219 |
+
"learning_rate": 0.001,
|
| 220 |
+
"loss": 16.765740966796876,
|
| 221 |
+
"step": 420
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 0.23726071717444056,
|
| 225 |
+
"grad_norm": 16.625,
|
| 226 |
+
"learning_rate": 0.001,
|
| 227 |
+
"loss": 16.70984344482422,
|
| 228 |
+
"step": 440
|
| 229 |
+
},
|
| 230 |
+
{
|
| 231 |
+
"epoch": 0.2426530062011324,
|
| 232 |
+
"eval_loss": 4.115875244140625,
|
| 233 |
+
"eval_runtime": 17.5037,
|
| 234 |
+
"eval_samples_per_second": 544.283,
|
| 235 |
+
"eval_steps_per_second": 4.285,
|
| 236 |
+
"step": 450
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.2480452952278242,
|
| 240 |
+
"grad_norm": 9.625,
|
| 241 |
+
"learning_rate": 0.001,
|
| 242 |
+
"loss": 16.492950439453125,
|
| 243 |
+
"step": 460
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"epoch": 0.2588298732812079,
|
| 247 |
+
"grad_norm": 4.4375,
|
| 248 |
+
"learning_rate": 0.001,
|
| 249 |
+
"loss": 16.204039001464842,
|
| 250 |
+
"step": 480
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.26961445133459155,
|
| 254 |
+
"grad_norm": 3.21875,
|
| 255 |
+
"learning_rate": 0.001,
|
| 256 |
+
"loss": 15.97710723876953,
|
| 257 |
+
"step": 500
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"epoch": 0.26961445133459155,
|
| 261 |
+
"eval_loss": 3.9576539993286133,
|
| 262 |
+
"eval_runtime": 16.9826,
|
| 263 |
+
"eval_samples_per_second": 560.986,
|
| 264 |
+
"eval_steps_per_second": 4.416,
|
| 265 |
+
"step": 500
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.2803990293879752,
|
| 269 |
+
"grad_norm": 10.5,
|
| 270 |
+
"learning_rate": 0.001,
|
| 271 |
+
"loss": 15.733917236328125,
|
| 272 |
+
"step": 520
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.29118360744135885,
|
| 276 |
+
"grad_norm": 9.8125,
|
| 277 |
+
"learning_rate": 0.001,
|
| 278 |
+
"loss": 15.579618835449219,
|
| 279 |
+
"step": 540
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.2965758964680507,
|
| 283 |
+
"eval_loss": 3.861560821533203,
|
| 284 |
+
"eval_runtime": 16.9704,
|
| 285 |
+
"eval_samples_per_second": 561.389,
|
| 286 |
+
"eval_steps_per_second": 4.419,
|
| 287 |
+
"step": 550
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"epoch": 0.3019681854947425,
|
| 291 |
+
"grad_norm": 3.984375,
|
| 292 |
+
"learning_rate": 0.001,
|
| 293 |
+
"loss": 15.382864379882813,
|
| 294 |
+
"step": 560
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.3127527635481262,
|
| 298 |
+
"grad_norm": 2.75,
|
| 299 |
+
"learning_rate": 0.001,
|
| 300 |
+
"loss": 15.098277282714843,
|
| 301 |
+
"step": 580
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"epoch": 0.3235373416015098,
|
| 305 |
+
"grad_norm": 5.75,
|
| 306 |
+
"learning_rate": 0.001,
|
| 307 |
+
"loss": 14.87418212890625,
|
| 308 |
+
"step": 600
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"epoch": 0.3235373416015098,
|
| 312 |
+
"eval_loss": 3.683532238006592,
|
| 313 |
+
"eval_runtime": 17.1359,
|
| 314 |
+
"eval_samples_per_second": 555.968,
|
| 315 |
+
"eval_steps_per_second": 4.377,
|
| 316 |
+
"step": 600
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"epoch": 0.3343219196548935,
|
| 320 |
+
"grad_norm": 5.28125,
|
| 321 |
+
"learning_rate": 0.001,
|
| 322 |
+
"loss": 14.851046752929687,
|
| 323 |
+
"step": 620
|
| 324 |
+
},
|
| 325 |
+
{
|
| 326 |
+
"epoch": 0.34510649770827717,
|
| 327 |
+
"grad_norm": 5.125,
|
| 328 |
+
"learning_rate": 0.001,
|
| 329 |
+
"loss": 14.559681701660157,
|
| 330 |
+
"step": 640
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.350498786734969,
|
| 334 |
+
"eval_loss": 3.5849404335021973,
|
| 335 |
+
"eval_runtime": 16.9858,
|
| 336 |
+
"eval_samples_per_second": 560.88,
|
| 337 |
+
"eval_steps_per_second": 4.415,
|
| 338 |
+
"step": 650
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.35589107576166085,
|
| 342 |
+
"grad_norm": 7.375,
|
| 343 |
+
"learning_rate": 0.001,
|
| 344 |
+
"loss": 14.377513122558593,
|
| 345 |
+
"step": 660
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.36667565381504447,
|
| 349 |
+
"grad_norm": 4.0625,
|
| 350 |
+
"learning_rate": 0.001,
|
| 351 |
+
"loss": 14.217088317871093,
|
| 352 |
+
"step": 680
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.37746023186842814,
|
| 356 |
+
"grad_norm": 5.0625,
|
| 357 |
+
"learning_rate": 0.001,
|
| 358 |
+
"loss": 14.131674194335938,
|
| 359 |
+
"step": 700
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.37746023186842814,
|
| 363 |
+
"eval_loss": 3.5239312648773193,
|
| 364 |
+
"eval_runtime": 16.9182,
|
| 365 |
+
"eval_samples_per_second": 563.122,
|
| 366 |
+
"eval_steps_per_second": 4.433,
|
| 367 |
+
"step": 700
|
| 368 |
+
},
|
| 369 |
+
{
|
| 370 |
+
"epoch": 0.3882448099218118,
|
| 371 |
+
"grad_norm": 6.59375,
|
| 372 |
+
"learning_rate": 0.001,
|
| 373 |
+
"loss": 13.94341583251953,
|
| 374 |
+
"step": 720
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 0.3990293879751955,
|
| 378 |
+
"grad_norm": 4.0625,
|
| 379 |
+
"learning_rate": 0.001,
|
| 380 |
+
"loss": 13.774238586425781,
|
| 381 |
+
"step": 740
|
| 382 |
+
},
|
| 383 |
+
{
|
| 384 |
+
"epoch": 0.4044216770018873,
|
| 385 |
+
"eval_loss": 3.419768810272217,
|
| 386 |
+
"eval_runtime": 17.1125,
|
| 387 |
+
"eval_samples_per_second": 556.727,
|
| 388 |
+
"eval_steps_per_second": 4.383,
|
| 389 |
+
"step": 750
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.4098139660285791,
|
| 393 |
+
"grad_norm": 4.09375,
|
| 394 |
+
"learning_rate": 0.001,
|
| 395 |
+
"loss": 13.6376220703125,
|
| 396 |
+
"step": 760
|
| 397 |
+
},
|
| 398 |
+
{
|
| 399 |
+
"epoch": 0.4205985440819628,
|
| 400 |
+
"grad_norm": 9.0625,
|
| 401 |
+
"learning_rate": 0.001,
|
| 402 |
+
"loss": 13.435893249511718,
|
| 403 |
+
"step": 780
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 0.43138312213534646,
|
| 407 |
+
"grad_norm": 2.28125,
|
| 408 |
+
"learning_rate": 0.001,
|
| 409 |
+
"loss": 13.210621643066407,
|
| 410 |
+
"step": 800
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.43138312213534646,
|
| 414 |
+
"eval_loss": 3.365565776824951,
|
| 415 |
+
"eval_runtime": 17.0151,
|
| 416 |
+
"eval_samples_per_second": 559.916,
|
| 417 |
+
"eval_steps_per_second": 4.408,
|
| 418 |
+
"step": 800
|
| 419 |
+
}
|
| 420 |
+
],
|
| 421 |
+
"logging_steps": 20,
|
| 422 |
+
"max_steps": 1500,
|
| 423 |
+
"num_input_tokens_seen": 0,
|
| 424 |
+
"num_train_epochs": 1,
|
| 425 |
+
"save_steps": 100,
|
| 426 |
+
"stateful_callbacks": {
|
| 427 |
+
"TrainerControl": {
|
| 428 |
+
"args": {
|
| 429 |
+
"should_epoch_stop": false,
|
| 430 |
+
"should_evaluate": false,
|
| 431 |
+
"should_log": false,
|
| 432 |
+
"should_save": true,
|
| 433 |
+
"should_training_stop": false
|
| 434 |
+
},
|
| 435 |
+
"attributes": {}
|
| 436 |
+
}
|
| 437 |
+
},
|
| 438 |
+
"total_flos": 1.94093330202624e+16,
|
| 439 |
+
"train_batch_size": 128,
|
| 440 |
+
"trial_name": null,
|
| 441 |
+
"trial_params": null
|
| 442 |
+
}
|
out/glu-waleed-94L_run/checkpoint-800/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e12a35f98937379bc279e337a2141ffbc223c6273ca4783d094a415f227e198
|
| 3 |
+
size 4920
|
out/glu-waleed-94L_run/checkpoint-900/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "waleed",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/glu-waleed-94L_run/checkpoint-900/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b00a6c9ac4860fcc186e1a6d25783cfbbdecda51d90da7c095ab0d2c847fdfbf
|
| 3 |
+
size 31992520
|
out/glu-waleed-94L_run/checkpoint-900/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6143e05ffce71c7b89b8e4ba08fcf2f97a619dee12eaf1409ba1b6ba90da8f93
|
| 3 |
+
size 64520514
|
out/glu-waleed-94L_run/checkpoint-900/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:86a2aa7bca84d2b8e7dd4e04a286714ba9169af11c46c739950a52df4c45259f
|
| 3 |
+
size 14244
|
out/glu-waleed-94L_run/checkpoint-900/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6d48b00fea613f19335b87fb622648d28d7f8900963d07ba43c75df2b4dc3d4e
|
| 3 |
+
size 1064
|
out/glu-waleed-94L_run/checkpoint-900/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/glu-waleed-94L_run/checkpoint-900/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/glu-waleed-94L_run/checkpoint-900/trainer_state.json
ADDED
|
@@ -0,0 +1,493 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.4853060124022648,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 900,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.625,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.4543212890625,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.2314453125,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.255674743652342,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.911471843719482,
|
| 29 |
+
"eval_runtime": 20.9396,
|
| 30 |
+
"eval_samples_per_second": 454.975,
|
| 31 |
+
"eval_steps_per_second": 3.582,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 8.75,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.663941955566408,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 4.15625,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 23.1123046875,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 1.578125,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.97301483154297,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.725912094116211,
|
| 58 |
+
"eval_runtime": 20.3952,
|
| 59 |
+
"eval_samples_per_second": 467.12,
|
| 60 |
+
"eval_steps_per_second": 3.677,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 0.2099609375,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 22.912661743164062,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 0.859375,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 22.89833984375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.71552038192749,
|
| 80 |
+
"eval_runtime": 20.5186,
|
| 81 |
+
"eval_samples_per_second": 464.31,
|
| 82 |
+
"eval_steps_per_second": 3.655,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 5.3125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 22.871157836914062,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 3.515625,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 22.64031982421875,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 2.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 21.882586669921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 5.379149913787842,
|
| 109 |
+
"eval_runtime": 19.954,
|
| 110 |
+
"eval_samples_per_second": 477.447,
|
| 111 |
+
"eval_steps_per_second": 3.759,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.11863035858722028,
|
| 116 |
+
"grad_norm": 3.0,
|
| 117 |
+
"learning_rate": 0.001,
|
| 118 |
+
"loss": 21.073570251464844,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.12941493664060394,
|
| 123 |
+
"grad_norm": 26.625,
|
| 124 |
+
"learning_rate": 0.001,
|
| 125 |
+
"loss": 20.38078918457031,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.13480722566729578,
|
| 130 |
+
"eval_loss": 4.969682693481445,
|
| 131 |
+
"eval_runtime": 20.8528,
|
| 132 |
+
"eval_samples_per_second": 456.87,
|
| 133 |
+
"eval_steps_per_second": 3.597,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.1401995146939876,
|
| 138 |
+
"grad_norm": 5.5625,
|
| 139 |
+
"learning_rate": 0.001,
|
| 140 |
+
"loss": 19.993435668945313,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.15098409274737126,
|
| 145 |
+
"grad_norm": 3.84375,
|
| 146 |
+
"learning_rate": 0.001,
|
| 147 |
+
"loss": 19.312832641601563,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.1617686708007549,
|
| 152 |
+
"grad_norm": 8.1875,
|
| 153 |
+
"learning_rate": 0.001,
|
| 154 |
+
"loss": 18.880908203125,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.1617686708007549,
|
| 159 |
+
"eval_loss": 4.66710090637207,
|
| 160 |
+
"eval_runtime": 20.0053,
|
| 161 |
+
"eval_samples_per_second": 476.224,
|
| 162 |
+
"eval_steps_per_second": 3.749,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.17255324885413859,
|
| 167 |
+
"grad_norm": 8.6875,
|
| 168 |
+
"learning_rate": 0.001,
|
| 169 |
+
"loss": 18.437432861328126,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.18333782690752223,
|
| 174 |
+
"grad_norm": 8.0,
|
| 175 |
+
"learning_rate": 0.001,
|
| 176 |
+
"loss": 18.02788543701172,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.18873011593421407,
|
| 181 |
+
"eval_loss": 4.424458980560303,
|
| 182 |
+
"eval_runtime": 20.6161,
|
| 183 |
+
"eval_samples_per_second": 462.114,
|
| 184 |
+
"eval_steps_per_second": 3.638,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.1941224049609059,
|
| 189 |
+
"grad_norm": 9.9375,
|
| 190 |
+
"learning_rate": 0.001,
|
| 191 |
+
"loss": 17.731655883789063,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.20490698301428956,
|
| 196 |
+
"grad_norm": 15.625,
|
| 197 |
+
"learning_rate": 0.001,
|
| 198 |
+
"loss": 17.360447692871094,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.21569156106767323,
|
| 203 |
+
"grad_norm": 4.71875,
|
| 204 |
+
"learning_rate": 0.001,
|
| 205 |
+
"loss": 17.05499267578125,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.21569156106767323,
|
| 210 |
+
"eval_loss": 4.228377342224121,
|
| 211 |
+
"eval_runtime": 20.6518,
|
| 212 |
+
"eval_samples_per_second": 461.316,
|
| 213 |
+
"eval_steps_per_second": 3.632,
|
| 214 |
+
"step": 400
|
| 215 |
+
},
|
| 216 |
+
{
|
| 217 |
+
"epoch": 0.22647613912105688,
|
| 218 |
+
"grad_norm": 3.03125,
|
| 219 |
+
"learning_rate": 0.001,
|
| 220 |
+
"loss": 16.765740966796876,
|
| 221 |
+
"step": 420
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 0.23726071717444056,
|
| 225 |
+
"grad_norm": 16.625,
|
| 226 |
+
"learning_rate": 0.001,
|
| 227 |
+
"loss": 16.70984344482422,
|
| 228 |
+
"step": 440
|
| 229 |
+
},
|
| 230 |
+
{
|
| 231 |
+
"epoch": 0.2426530062011324,
|
| 232 |
+
"eval_loss": 4.115875244140625,
|
| 233 |
+
"eval_runtime": 17.5037,
|
| 234 |
+
"eval_samples_per_second": 544.283,
|
| 235 |
+
"eval_steps_per_second": 4.285,
|
| 236 |
+
"step": 450
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.2480452952278242,
|
| 240 |
+
"grad_norm": 9.625,
|
| 241 |
+
"learning_rate": 0.001,
|
| 242 |
+
"loss": 16.492950439453125,
|
| 243 |
+
"step": 460
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"epoch": 0.2588298732812079,
|
| 247 |
+
"grad_norm": 4.4375,
|
| 248 |
+
"learning_rate": 0.001,
|
| 249 |
+
"loss": 16.204039001464842,
|
| 250 |
+
"step": 480
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.26961445133459155,
|
| 254 |
+
"grad_norm": 3.21875,
|
| 255 |
+
"learning_rate": 0.001,
|
| 256 |
+
"loss": 15.97710723876953,
|
| 257 |
+
"step": 500
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"epoch": 0.26961445133459155,
|
| 261 |
+
"eval_loss": 3.9576539993286133,
|
| 262 |
+
"eval_runtime": 16.9826,
|
| 263 |
+
"eval_samples_per_second": 560.986,
|
| 264 |
+
"eval_steps_per_second": 4.416,
|
| 265 |
+
"step": 500
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.2803990293879752,
|
| 269 |
+
"grad_norm": 10.5,
|
| 270 |
+
"learning_rate": 0.001,
|
| 271 |
+
"loss": 15.733917236328125,
|
| 272 |
+
"step": 520
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.29118360744135885,
|
| 276 |
+
"grad_norm": 9.8125,
|
| 277 |
+
"learning_rate": 0.001,
|
| 278 |
+
"loss": 15.579618835449219,
|
| 279 |
+
"step": 540
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.2965758964680507,
|
| 283 |
+
"eval_loss": 3.861560821533203,
|
| 284 |
+
"eval_runtime": 16.9704,
|
| 285 |
+
"eval_samples_per_second": 561.389,
|
| 286 |
+
"eval_steps_per_second": 4.419,
|
| 287 |
+
"step": 550
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"epoch": 0.3019681854947425,
|
| 291 |
+
"grad_norm": 3.984375,
|
| 292 |
+
"learning_rate": 0.001,
|
| 293 |
+
"loss": 15.382864379882813,
|
| 294 |
+
"step": 560
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.3127527635481262,
|
| 298 |
+
"grad_norm": 2.75,
|
| 299 |
+
"learning_rate": 0.001,
|
| 300 |
+
"loss": 15.098277282714843,
|
| 301 |
+
"step": 580
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"epoch": 0.3235373416015098,
|
| 305 |
+
"grad_norm": 5.75,
|
| 306 |
+
"learning_rate": 0.001,
|
| 307 |
+
"loss": 14.87418212890625,
|
| 308 |
+
"step": 600
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"epoch": 0.3235373416015098,
|
| 312 |
+
"eval_loss": 3.683532238006592,
|
| 313 |
+
"eval_runtime": 17.1359,
|
| 314 |
+
"eval_samples_per_second": 555.968,
|
| 315 |
+
"eval_steps_per_second": 4.377,
|
| 316 |
+
"step": 600
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"epoch": 0.3343219196548935,
|
| 320 |
+
"grad_norm": 5.28125,
|
| 321 |
+
"learning_rate": 0.001,
|
| 322 |
+
"loss": 14.851046752929687,
|
| 323 |
+
"step": 620
|
| 324 |
+
},
|
| 325 |
+
{
|
| 326 |
+
"epoch": 0.34510649770827717,
|
| 327 |
+
"grad_norm": 5.125,
|
| 328 |
+
"learning_rate": 0.001,
|
| 329 |
+
"loss": 14.559681701660157,
|
| 330 |
+
"step": 640
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.350498786734969,
|
| 334 |
+
"eval_loss": 3.5849404335021973,
|
| 335 |
+
"eval_runtime": 16.9858,
|
| 336 |
+
"eval_samples_per_second": 560.88,
|
| 337 |
+
"eval_steps_per_second": 4.415,
|
| 338 |
+
"step": 650
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.35589107576166085,
|
| 342 |
+
"grad_norm": 7.375,
|
| 343 |
+
"learning_rate": 0.001,
|
| 344 |
+
"loss": 14.377513122558593,
|
| 345 |
+
"step": 660
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.36667565381504447,
|
| 349 |
+
"grad_norm": 4.0625,
|
| 350 |
+
"learning_rate": 0.001,
|
| 351 |
+
"loss": 14.217088317871093,
|
| 352 |
+
"step": 680
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.37746023186842814,
|
| 356 |
+
"grad_norm": 5.0625,
|
| 357 |
+
"learning_rate": 0.001,
|
| 358 |
+
"loss": 14.131674194335938,
|
| 359 |
+
"step": 700
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.37746023186842814,
|
| 363 |
+
"eval_loss": 3.5239312648773193,
|
| 364 |
+
"eval_runtime": 16.9182,
|
| 365 |
+
"eval_samples_per_second": 563.122,
|
| 366 |
+
"eval_steps_per_second": 4.433,
|
| 367 |
+
"step": 700
|
| 368 |
+
},
|
| 369 |
+
{
|
| 370 |
+
"epoch": 0.3882448099218118,
|
| 371 |
+
"grad_norm": 6.59375,
|
| 372 |
+
"learning_rate": 0.001,
|
| 373 |
+
"loss": 13.94341583251953,
|
| 374 |
+
"step": 720
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 0.3990293879751955,
|
| 378 |
+
"grad_norm": 4.0625,
|
| 379 |
+
"learning_rate": 0.001,
|
| 380 |
+
"loss": 13.774238586425781,
|
| 381 |
+
"step": 740
|
| 382 |
+
},
|
| 383 |
+
{
|
| 384 |
+
"epoch": 0.4044216770018873,
|
| 385 |
+
"eval_loss": 3.419768810272217,
|
| 386 |
+
"eval_runtime": 17.1125,
|
| 387 |
+
"eval_samples_per_second": 556.727,
|
| 388 |
+
"eval_steps_per_second": 4.383,
|
| 389 |
+
"step": 750
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.4098139660285791,
|
| 393 |
+
"grad_norm": 4.09375,
|
| 394 |
+
"learning_rate": 0.001,
|
| 395 |
+
"loss": 13.6376220703125,
|
| 396 |
+
"step": 760
|
| 397 |
+
},
|
| 398 |
+
{
|
| 399 |
+
"epoch": 0.4205985440819628,
|
| 400 |
+
"grad_norm": 9.0625,
|
| 401 |
+
"learning_rate": 0.001,
|
| 402 |
+
"loss": 13.435893249511718,
|
| 403 |
+
"step": 780
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 0.43138312213534646,
|
| 407 |
+
"grad_norm": 2.28125,
|
| 408 |
+
"learning_rate": 0.001,
|
| 409 |
+
"loss": 13.210621643066407,
|
| 410 |
+
"step": 800
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.43138312213534646,
|
| 414 |
+
"eval_loss": 3.365565776824951,
|
| 415 |
+
"eval_runtime": 17.0151,
|
| 416 |
+
"eval_samples_per_second": 559.916,
|
| 417 |
+
"eval_steps_per_second": 4.408,
|
| 418 |
+
"step": 800
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.44216770018873014,
|
| 422 |
+
"grad_norm": 5.125,
|
| 423 |
+
"learning_rate": 0.001,
|
| 424 |
+
"loss": 13.199737548828125,
|
| 425 |
+
"step": 820
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.45295227824211376,
|
| 429 |
+
"grad_norm": 2.671875,
|
| 430 |
+
"learning_rate": 0.001,
|
| 431 |
+
"loss": 12.949116516113282,
|
| 432 |
+
"step": 840
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.4583445672688056,
|
| 436 |
+
"eval_loss": 3.1935510635375977,
|
| 437 |
+
"eval_runtime": 17.0147,
|
| 438 |
+
"eval_samples_per_second": 559.928,
|
| 439 |
+
"eval_steps_per_second": 4.408,
|
| 440 |
+
"step": 850
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.46373685629549743,
|
| 444 |
+
"grad_norm": 2.6875,
|
| 445 |
+
"learning_rate": 0.001,
|
| 446 |
+
"loss": 12.769676208496094,
|
| 447 |
+
"step": 860
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.4745214343488811,
|
| 451 |
+
"grad_norm": 2.8125,
|
| 452 |
+
"learning_rate": 0.001,
|
| 453 |
+
"loss": 12.551689147949219,
|
| 454 |
+
"step": 880
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.4853060124022648,
|
| 458 |
+
"grad_norm": 17.125,
|
| 459 |
+
"learning_rate": 0.001,
|
| 460 |
+
"loss": 12.372489166259765,
|
| 461 |
+
"step": 900
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.4853060124022648,
|
| 465 |
+
"eval_loss": 3.1010639667510986,
|
| 466 |
+
"eval_runtime": 17.1354,
|
| 467 |
+
"eval_samples_per_second": 555.982,
|
| 468 |
+
"eval_steps_per_second": 4.377,
|
| 469 |
+
"step": 900
|
| 470 |
+
}
|
| 471 |
+
],
|
| 472 |
+
"logging_steps": 20,
|
| 473 |
+
"max_steps": 1500,
|
| 474 |
+
"num_input_tokens_seen": 0,
|
| 475 |
+
"num_train_epochs": 1,
|
| 476 |
+
"save_steps": 100,
|
| 477 |
+
"stateful_callbacks": {
|
| 478 |
+
"TrainerControl": {
|
| 479 |
+
"args": {
|
| 480 |
+
"should_epoch_stop": false,
|
| 481 |
+
"should_evaluate": false,
|
| 482 |
+
"should_log": false,
|
| 483 |
+
"should_save": true,
|
| 484 |
+
"should_training_stop": false
|
| 485 |
+
},
|
| 486 |
+
"attributes": {}
|
| 487 |
+
}
|
| 488 |
+
},
|
| 489 |
+
"total_flos": 2.18354996477952e+16,
|
| 490 |
+
"train_batch_size": 128,
|
| 491 |
+
"trial_name": null,
|
| 492 |
+
"trial_params": null
|
| 493 |
+
}
|
out/glu-waleed-94L_run/checkpoint-900/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e12a35f98937379bc279e337a2141ffbc223c6273ca4783d094a415f227e198
|
| 3 |
+
size 4920
|
out/glu-waleed-94L_run/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "waleed",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/glu-waleed-94L_run/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:390a40e963225f57b23ca14d6aeac4bd92e7d8e5d4791f0724adfa827128196a
|
| 3 |
+
size 31992520
|
out/glu-waleed-94L_run/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/glu-waleed-94L_run/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/glu-waleed-94L_run/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e12a35f98937379bc279e337a2141ffbc223c6273ca4783d094a415f227e198
|
| 3 |
+
size 4920
|
out/glu-waleed-94L_run/training_log.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/mlp-s10-94L_run/README.md
ADDED
|
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
tags:
|
| 4 |
+
- generated_from_trainer
|
| 5 |
+
model-index:
|
| 6 |
+
- name: A-mlp-s10-94L
|
| 7 |
+
results: []
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 11 |
+
should probably proofread and complete it, then remove this comment. -->
|
| 12 |
+
|
| 13 |
+
# A-mlp-s10-94L
|
| 14 |
+
|
| 15 |
+
This model is a fine-tuned version of [](https://huggingface.co/) on an unknown dataset.
|
| 16 |
+
It achieves the following results on the evaluation set:
|
| 17 |
+
- Loss: 1.8618
|
| 18 |
+
|
| 19 |
+
## Model description
|
| 20 |
+
|
| 21 |
+
More information needed
|
| 22 |
+
|
| 23 |
+
## Intended uses & limitations
|
| 24 |
+
|
| 25 |
+
More information needed
|
| 26 |
+
|
| 27 |
+
## Training and evaluation data
|
| 28 |
+
|
| 29 |
+
More information needed
|
| 30 |
+
|
| 31 |
+
## Training procedure
|
| 32 |
+
|
| 33 |
+
### Training hyperparameters
|
| 34 |
+
|
| 35 |
+
The following hyperparameters were used during training:
|
| 36 |
+
- learning_rate: 0.001
|
| 37 |
+
- train_batch_size: 128
|
| 38 |
+
- eval_batch_size: 128
|
| 39 |
+
- seed: 42
|
| 40 |
+
- gradient_accumulation_steps: 4
|
| 41 |
+
- total_train_batch_size: 512
|
| 42 |
+
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
| 43 |
+
- lr_scheduler_type: constant
|
| 44 |
+
- training_steps: 1500
|
| 45 |
+
|
| 46 |
+
### Training results
|
| 47 |
+
|
| 48 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
| 49 |
+
|:-------------:|:------:|:----:|:---------------:|
|
| 50 |
+
| 24.2798 | 0.0270 | 50 | 5.9086 |
|
| 51 |
+
| 22.1371 | 0.0539 | 100 | 5.4433 |
|
| 52 |
+
| 20.9160 | 0.0809 | 150 | 5.0314 |
|
| 53 |
+
| 18.3162 | 0.1078 | 200 | 4.4591 |
|
| 54 |
+
| 16.8577 | 0.1348 | 250 | 4.0378 |
|
| 55 |
+
| 15.2626 | 0.1618 | 300 | 3.7790 |
|
| 56 |
+
| 14.4196 | 0.1887 | 350 | 3.5089 |
|
| 57 |
+
| 13.3537 | 0.2157 | 400 | 3.3230 |
|
| 58 |
+
| 12.7840 | 0.2427 | 450 | 3.1206 |
|
| 59 |
+
| 11.9211 | 0.2696 | 500 | 2.9508 |
|
| 60 |
+
| 11.4566 | 0.2966 | 550 | 2.8138 |
|
| 61 |
+
| 10.8289 | 0.3235 | 600 | 2.6942 |
|
| 62 |
+
| 10.4861 | 0.3505 | 650 | 2.5830 |
|
| 63 |
+
| 10.0049 | 0.3775 | 700 | 2.4851 |
|
| 64 |
+
| 9.7264 | 0.4044 | 750 | 2.3961 |
|
| 65 |
+
| 9.3039 | 0.4314 | 800 | 2.3193 |
|
| 66 |
+
| 9.1100 | 0.4583 | 850 | 2.2558 |
|
| 67 |
+
| 8.7980 | 0.4853 | 900 | 2.1991 |
|
| 68 |
+
| 8.6568 | 0.5123 | 950 | 2.1517 |
|
| 69 |
+
| 8.4551 | 0.5392 | 1000 | 2.1050 |
|
| 70 |
+
| 8.2907 | 0.5662 | 1050 | 2.0706 |
|
| 71 |
+
| 8.1491 | 0.5932 | 1100 | 2.0383 |
|
| 72 |
+
| 8.0428 | 0.6201 | 1150 | 2.0064 |
|
| 73 |
+
| 7.8927 | 0.6471 | 1200 | 1.9802 |
|
| 74 |
+
| 7.8279 | 0.6740 | 1250 | 1.9556 |
|
| 75 |
+
| 7.7145 | 0.7010 | 1300 | 1.9351 |
|
| 76 |
+
| 7.6557 | 0.7280 | 1350 | 1.9123 |
|
| 77 |
+
| 7.5586 | 0.7549 | 1400 | 1.8938 |
|
| 78 |
+
| 7.5087 | 0.7819 | 1450 | 1.8763 |
|
| 79 |
+
| 7.4194 | 0.8088 | 1500 | 1.8618 |
|
| 80 |
+
|
| 81 |
+
|
| 82 |
+
### Framework versions
|
| 83 |
+
|
| 84 |
+
- Transformers 5.15.0.dev0
|
| 85 |
+
- Pytorch 2.6.0+cu124
|
| 86 |
+
- Datasets 5.0.1
|
| 87 |
+
- Tokenizers 0.22.2
|
out/mlp-s10-94L_run/checkpoint-100/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/mlp-s10-94L_run/checkpoint-100/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2be26509aa8682be1562cdefdc4fb702fa74f487c6ebeb27a02060b133a0c94a
|
| 3 |
+
size 31981960
|
out/mlp-s10-94L_run/checkpoint-100/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:de4ee11bbfdf5bd8a135bcd6b03e23684bb89f9318b0733ff65ec4481b31abe2
|
| 3 |
+
size 64439848
|
out/mlp-s10-94L_run/checkpoint-100/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3cf9097d4513154245c48236b6ec5137b7ee2a21c9f58f2cba798ea275c6026f
|
| 3 |
+
size 14244
|
out/mlp-s10-94L_run/checkpoint-100/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:906745ab5f61e73e8c1ba850ef3b839e2766eba889b49c843836266988895165
|
| 3 |
+
size 1064
|
out/mlp-s10-94L_run/checkpoint-100/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/mlp-s10-94L_run/checkpoint-100/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/mlp-s10-94L_run/checkpoint-100/trainer_state.json
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.05392289026691831,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 100,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.75,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.570291137695314,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.263671875,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.279835510253907,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.908648490905762,
|
| 29 |
+
"eval_runtime": 16.3836,
|
| 30 |
+
"eval_samples_per_second": 581.495,
|
| 31 |
+
"eval_steps_per_second": 4.578,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 2.03125,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.61173858642578,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 3.9375,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 22.80414733886719,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 4.5,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.137115478515625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.443320274353027,
|
| 58 |
+
"eval_runtime": 16.3492,
|
| 59 |
+
"eval_samples_per_second": 582.72,
|
| 60 |
+
"eval_steps_per_second": 4.587,
|
| 61 |
+
"step": 100
|
| 62 |
+
}
|
| 63 |
+
],
|
| 64 |
+
"logging_steps": 20,
|
| 65 |
+
"max_steps": 1500,
|
| 66 |
+
"num_input_tokens_seen": 0,
|
| 67 |
+
"num_train_epochs": 1,
|
| 68 |
+
"save_steps": 100,
|
| 69 |
+
"stateful_callbacks": {
|
| 70 |
+
"TrainerControl": {
|
| 71 |
+
"args": {
|
| 72 |
+
"should_epoch_stop": false,
|
| 73 |
+
"should_evaluate": false,
|
| 74 |
+
"should_log": false,
|
| 75 |
+
"should_save": true,
|
| 76 |
+
"should_training_stop": false
|
| 77 |
+
},
|
| 78 |
+
"attributes": {}
|
| 79 |
+
}
|
| 80 |
+
},
|
| 81 |
+
"total_flos": 2426166627532800.0,
|
| 82 |
+
"train_batch_size": 128,
|
| 83 |
+
"trial_name": null,
|
| 84 |
+
"trial_params": null
|
| 85 |
+
}
|
out/mlp-s10-94L_run/checkpoint-100/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:873bf272a464f8d4f32a1497a17e4dd88aaf0d116e87d1e95dbff747318696f0
|
| 3 |
+
size 4920
|
out/mlp-s10-94L_run/checkpoint-1000/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|