Upload training_log_25pct_step1656.json with huggingface_hub
Browse files- training_log_25pct_step1656.json +233 -0
training_log_25pct_step1656.json
ADDED
|
@@ -0,0 +1,233 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config": {
|
| 3 |
+
"epochs": 1,
|
| 4 |
+
"batch_size": 32,
|
| 5 |
+
"gradient_accumulation_steps": 4,
|
| 6 |
+
"effective_batch_size": 128,
|
| 7 |
+
"cross_max_length": 64,
|
| 8 |
+
"lora_r": 8,
|
| 9 |
+
"lora_alpha": 16,
|
| 10 |
+
"target_modules": [
|
| 11 |
+
"query_key_value"
|
| 12 |
+
],
|
| 13 |
+
"total_steps": 6625,
|
| 14 |
+
"start_time": "2026-01-11T02:43:56.393662"
|
| 15 |
+
},
|
| 16 |
+
"steps": [
|
| 17 |
+
{
|
| 18 |
+
"step": 100,
|
| 19 |
+
"epoch": 0,
|
| 20 |
+
"loss": 3.048646457195282,
|
| 21 |
+
"perplexity": 21.08678323378463,
|
| 22 |
+
"grad_norm": 0.9999998427911615,
|
| 23 |
+
"lr": 3.7764350453172205e-06,
|
| 24 |
+
"cross_attn_scale": 0.1510574018126888,
|
| 25 |
+
"tokens_per_sec": 14120.758195630631,
|
| 26 |
+
"step_time_ms": 404.4312286376953,
|
| 27 |
+
"elapsed_sec": 48.608083963394165,
|
| 28 |
+
"progress_pct": 1.509433962264151
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"step": 200,
|
| 32 |
+
"epoch": 0,
|
| 33 |
+
"loss": 3.0407689785957337,
|
| 34 |
+
"perplexity": 20.921325102192466,
|
| 35 |
+
"grad_norm": 0.9999998367539097,
|
| 36 |
+
"lr": 7.552870090634441e-06,
|
| 37 |
+
"cross_attn_scale": 0.3021148036253776,
|
| 38 |
+
"tokens_per_sec": 14170.247728402273,
|
| 39 |
+
"step_time_ms": 392.75057315826416,
|
| 40 |
+
"elapsed_sec": 96.21885418891907,
|
| 41 |
+
"progress_pct": 3.018867924528302
|
| 42 |
+
},
|
| 43 |
+
{
|
| 44 |
+
"step": 300,
|
| 45 |
+
"epoch": 0,
|
| 46 |
+
"loss": 3.032252535820007,
|
| 47 |
+
"perplexity": 20.743906394646302,
|
| 48 |
+
"grad_norm": 0.9999998386384089,
|
| 49 |
+
"lr": 1.1329305135951662e-05,
|
| 50 |
+
"cross_attn_scale": 0.45317220543806647,
|
| 51 |
+
"tokens_per_sec": 14193.356545724891,
|
| 52 |
+
"step_time_ms": 401.1916208267212,
|
| 53 |
+
"elapsed_sec": 144.4561047554016,
|
| 54 |
+
"progress_pct": 4.528301886792453
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 400,
|
| 58 |
+
"epoch": 0,
|
| 59 |
+
"loss": 3.007298858165741,
|
| 60 |
+
"perplexity": 20.232674724133254,
|
| 61 |
+
"grad_norm": 0.9999998386061594,
|
| 62 |
+
"lr": 1.5105740181268882e-05,
|
| 63 |
+
"cross_attn_scale": 0.6042296072507553,
|
| 64 |
+
"tokens_per_sec": 14183.614113827954,
|
| 65 |
+
"step_time_ms": 397.9473662376404,
|
| 66 |
+
"elapsed_sec": 192.37191438674927,
|
| 67 |
+
"progress_pct": 6.037735849056604
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"step": 500,
|
| 71 |
+
"epoch": 0,
|
| 72 |
+
"loss": 2.9540454363822937,
|
| 73 |
+
"perplexity": 19.183402199369795,
|
| 74 |
+
"grad_norm": 0.9999998439866878,
|
| 75 |
+
"lr": 1.8882175226586106e-05,
|
| 76 |
+
"cross_attn_scale": 0.7552870090634441,
|
| 77 |
+
"tokens_per_sec": 14180.037076479686,
|
| 78 |
+
"step_time_ms": 396.76615476608276,
|
| 79 |
+
"elapsed_sec": 240.06932997703552,
|
| 80 |
+
"progress_pct": 7.547169811320755
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"step": 600,
|
| 84 |
+
"epoch": 0,
|
| 85 |
+
"loss": 2.906618938446045,
|
| 86 |
+
"perplexity": 18.29483790655841,
|
| 87 |
+
"grad_norm": 0.9999998550880919,
|
| 88 |
+
"lr": 2.2658610271903323e-05,
|
| 89 |
+
"cross_attn_scale": 0.9063444108761329,
|
| 90 |
+
"tokens_per_sec": 14216.807056874239,
|
| 91 |
+
"step_time_ms": 393.05994033813477,
|
| 92 |
+
"elapsed_sec": 287.3540439605713,
|
| 93 |
+
"progress_pct": 9.056603773584905
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"step": 700,
|
| 97 |
+
"epoch": 0,
|
| 98 |
+
"loss": 2.8761880707740786,
|
| 99 |
+
"perplexity": 17.746495694600036,
|
| 100 |
+
"grad_norm": 0.9999998625671315,
|
| 101 |
+
"lr": 2.6435045317220547e-05,
|
| 102 |
+
"cross_attn_scale": 1,
|
| 103 |
+
"tokens_per_sec": 14246.662539076659,
|
| 104 |
+
"step_time_ms": 394.4803309440613,
|
| 105 |
+
"elapsed_sec": 334.8180661201477,
|
| 106 |
+
"progress_pct": 10.566037735849056
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"step": 800,
|
| 110 |
+
"epoch": 0,
|
| 111 |
+
"loss": 2.845837857723236,
|
| 112 |
+
"perplexity": 17.215977167189468,
|
| 113 |
+
"grad_norm": 0.9999998764891178,
|
| 114 |
+
"lr": 3.0211480362537764e-05,
|
| 115 |
+
"cross_attn_scale": 1,
|
| 116 |
+
"tokens_per_sec": 14252.739167051805,
|
| 117 |
+
"step_time_ms": 399.4402241706848,
|
| 118 |
+
"elapsed_sec": 382.8833837509155,
|
| 119 |
+
"progress_pct": 12.075471698113208
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"step": 900,
|
| 123 |
+
"epoch": 0,
|
| 124 |
+
"loss": 2.83284410238266,
|
| 125 |
+
"perplexity": 16.993724050291934,
|
| 126 |
+
"grad_norm": 0.9999998825152169,
|
| 127 |
+
"lr": 3.3987915407854985e-05,
|
| 128 |
+
"cross_attn_scale": 1,
|
| 129 |
+
"tokens_per_sec": 14264.806891693297,
|
| 130 |
+
"step_time_ms": 393.567271232605,
|
| 131 |
+
"elapsed_sec": 430.18689608573914,
|
| 132 |
+
"progress_pct": 13.584905660377359
|
| 133 |
+
},
|
| 134 |
+
{
|
| 135 |
+
"step": 1000,
|
| 136 |
+
"epoch": 0,
|
| 137 |
+
"loss": 2.8241940665245058,
|
| 138 |
+
"perplexity": 16.84736166045992,
|
| 139 |
+
"grad_norm": 0.9999998878949373,
|
| 140 |
+
"lr": 3.776435045317221e-05,
|
| 141 |
+
"cross_attn_scale": 1,
|
| 142 |
+
"tokens_per_sec": 14271.324496099065,
|
| 143 |
+
"step_time_ms": 391.49524688720703,
|
| 144 |
+
"elapsed_sec": 477.29774498939514,
|
| 145 |
+
"progress_pct": 15.09433962264151
|
| 146 |
+
},
|
| 147 |
+
{
|
| 148 |
+
"step": 1100,
|
| 149 |
+
"epoch": 0,
|
| 150 |
+
"loss": 2.801992883682251,
|
| 151 |
+
"perplexity": 16.47745171667062,
|
| 152 |
+
"grad_norm": 0.9999998981946228,
|
| 153 |
+
"lr": 4.1540785498489426e-05,
|
| 154 |
+
"cross_attn_scale": 1,
|
| 155 |
+
"tokens_per_sec": 14282.931722510191,
|
| 156 |
+
"step_time_ms": 391.6118621826172,
|
| 157 |
+
"elapsed_sec": 524.3816287517548,
|
| 158 |
+
"progress_pct": 16.60377358490566
|
| 159 |
+
},
|
| 160 |
+
{
|
| 161 |
+
"step": 1200,
|
| 162 |
+
"epoch": 0,
|
| 163 |
+
"loss": 2.7904515504837035,
|
| 164 |
+
"perplexity": 16.288373164237353,
|
| 165 |
+
"grad_norm": 0.9999999022882037,
|
| 166 |
+
"lr": 4.5317220543806646e-05,
|
| 167 |
+
"cross_attn_scale": 1,
|
| 168 |
+
"tokens_per_sec": 14296.256604794175,
|
| 169 |
+
"step_time_ms": 391.77978515625,
|
| 170 |
+
"elapsed_sec": 571.524786233902,
|
| 171 |
+
"progress_pct": 18.11320754716981
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 1300,
|
| 175 |
+
"epoch": 0,
|
| 176 |
+
"loss": 2.771877167224884,
|
| 177 |
+
"perplexity": 15.988619169285416,
|
| 178 |
+
"grad_norm": 0.9999999069481242,
|
| 179 |
+
"lr": 4.9093655589123874e-05,
|
| 180 |
+
"cross_attn_scale": 1,
|
| 181 |
+
"tokens_per_sec": 14298.154237716446,
|
| 182 |
+
"step_time_ms": 395.62140226364136,
|
| 183 |
+
"elapsed_sec": 619.1293542385101,
|
| 184 |
+
"progress_pct": 19.62264150943396
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"step": 1400,
|
| 188 |
+
"epoch": 0,
|
| 189 |
+
"loss": 2.75765634059906,
|
| 190 |
+
"perplexity": 15.762856856132169,
|
| 191 |
+
"grad_norm": 0.9999999069954555,
|
| 192 |
+
"lr": 5.2870090634441094e-05,
|
| 193 |
+
"cross_attn_scale": 1,
|
| 194 |
+
"tokens_per_sec": 14299.554305654967,
|
| 195 |
+
"step_time_ms": 394.1334104537964,
|
| 196 |
+
"elapsed_sec": 666.549795627594,
|
| 197 |
+
"progress_pct": 21.132075471698112
|
| 198 |
+
},
|
| 199 |
+
{
|
| 200 |
+
"step": 1500,
|
| 201 |
+
"epoch": 0,
|
| 202 |
+
"loss": 2.750157058238983,
|
| 203 |
+
"perplexity": 15.645088881345737,
|
| 204 |
+
"grad_norm": 0.9999999016993876,
|
| 205 |
+
"lr": 5.664652567975831e-05,
|
| 206 |
+
"cross_attn_scale": 1,
|
| 207 |
+
"tokens_per_sec": 14300.353267324579,
|
| 208 |
+
"step_time_ms": 393.2357144355774,
|
| 209 |
+
"elapsed_sec": 713.8556516170502,
|
| 210 |
+
"progress_pct": 22.641509433962263
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"step": 1600,
|
| 214 |
+
"epoch": 0,
|
| 215 |
+
"loss": 2.740102276802063,
|
| 216 |
+
"perplexity": 15.488569136653092,
|
| 217 |
+
"grad_norm": 0.9999999006110072,
|
| 218 |
+
"lr": 6.042296072507553e-05,
|
| 219 |
+
"cross_attn_scale": 1,
|
| 220 |
+
"tokens_per_sec": 14304.826303474321,
|
| 221 |
+
"step_time_ms": 397.52734184265137,
|
| 222 |
+
"elapsed_sec": 761.6850962638855,
|
| 223 |
+
"progress_pct": 24.150943396226417
|
| 224 |
+
}
|
| 225 |
+
],
|
| 226 |
+
"checkpoint": {
|
| 227 |
+
"pct": 25,
|
| 228 |
+
"step": 1656,
|
| 229 |
+
"final_loss": 2.745757640644803,
|
| 230 |
+
"total_tokens": 11277800,
|
| 231 |
+
"total_time_sec": 788.4210357666016
|
| 232 |
+
}
|
| 233 |
+
}
|