Upload training_log_75pct_step4968.json with huggingface_hub
Browse files- training_log_75pct_step4968.json +662 -0
training_log_75pct_step4968.json
ADDED
|
@@ -0,0 +1,662 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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 |
+
"step": 1700,
|
| 227 |
+
"epoch": 0,
|
| 228 |
+
"loss": 2.752125091552734,
|
| 229 |
+
"perplexity": 15.675909255274465,
|
| 230 |
+
"grad_norm": 0.9999998985642324,
|
| 231 |
+
"lr": 6.419939577039276e-05,
|
| 232 |
+
"cross_attn_scale": 1,
|
| 233 |
+
"tokens_per_sec": 14232.942905024669,
|
| 234 |
+
"step_time_ms": 393.9875912666321,
|
| 235 |
+
"elapsed_sec": 813.2436192035675,
|
| 236 |
+
"progress_pct": 25.660377358490567
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"step": 1800,
|
| 240 |
+
"epoch": 0,
|
| 241 |
+
"loss": 2.7508164858818054,
|
| 242 |
+
"perplexity": 15.65540908776923,
|
| 243 |
+
"grad_norm": 0.9999999009545788,
|
| 244 |
+
"lr": 6.797583081570997e-05,
|
| 245 |
+
"cross_attn_scale": 1,
|
| 246 |
+
"tokens_per_sec": 14238.47686208524,
|
| 247 |
+
"step_time_ms": 396.0763478279114,
|
| 248 |
+
"elapsed_sec": 860.9443354606628,
|
| 249 |
+
"progress_pct": 27.169811320754718
|
| 250 |
+
},
|
| 251 |
+
{
|
| 252 |
+
"step": 1900,
|
| 253 |
+
"epoch": 0,
|
| 254 |
+
"loss": 2.736833720207214,
|
| 255 |
+
"perplexity": 15.438026517559699,
|
| 256 |
+
"grad_norm": 0.9999998984068662,
|
| 257 |
+
"lr": 7.175226586102719e-05,
|
| 258 |
+
"cross_attn_scale": 1,
|
| 259 |
+
"tokens_per_sec": 14242.67026465119,
|
| 260 |
+
"step_time_ms": 393.61583709716797,
|
| 261 |
+
"elapsed_sec": 908.3369030952454,
|
| 262 |
+
"progress_pct": 28.67924528301887
|
| 263 |
+
},
|
| 264 |
+
{
|
| 265 |
+
"step": 2000,
|
| 266 |
+
"epoch": 0,
|
| 267 |
+
"loss": 2.7328391218185426,
|
| 268 |
+
"perplexity": 15.376480808742464,
|
| 269 |
+
"grad_norm": 0.9999998890564358,
|
| 270 |
+
"lr": 7.552870090634442e-05,
|
| 271 |
+
"cross_attn_scale": 1,
|
| 272 |
+
"tokens_per_sec": 14248.84505458971,
|
| 273 |
+
"step_time_ms": 396.22297763824463,
|
| 274 |
+
"elapsed_sec": 956.0212738513947,
|
| 275 |
+
"progress_pct": 30.18867924528302
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"step": 2100,
|
| 279 |
+
"epoch": 0,
|
| 280 |
+
"loss": 2.711907162666321,
|
| 281 |
+
"perplexity": 15.057966138185757,
|
| 282 |
+
"grad_norm": 0.9999998800023507,
|
| 283 |
+
"lr": 7.930513595166163e-05,
|
| 284 |
+
"cross_attn_scale": 1,
|
| 285 |
+
"tokens_per_sec": 14248.22051184832,
|
| 286 |
+
"step_time_ms": 394.12156105041504,
|
| 287 |
+
"elapsed_sec": 1003.8458478450775,
|
| 288 |
+
"progress_pct": 31.69811320754717
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 2200,
|
| 292 |
+
"epoch": 0,
|
| 293 |
+
"loss": 2.6982900667190552,
|
| 294 |
+
"perplexity": 14.854310117205573,
|
| 295 |
+
"grad_norm": 0.9999998720552749,
|
| 296 |
+
"lr": 8.308157099697885e-05,
|
| 297 |
+
"cross_attn_scale": 1,
|
| 298 |
+
"tokens_per_sec": 14253.967031040447,
|
| 299 |
+
"step_time_ms": 400.0247573852539,
|
| 300 |
+
"elapsed_sec": 1051.995628118515,
|
| 301 |
+
"progress_pct": 33.20754716981132
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"step": 2300,
|
| 305 |
+
"epoch": 0,
|
| 306 |
+
"loss": 2.6988302421569825,
|
| 307 |
+
"perplexity": 14.8623362182344,
|
| 308 |
+
"grad_norm": 0.9999998672299303,
|
| 309 |
+
"lr": 8.685800604229609e-05,
|
| 310 |
+
"cross_attn_scale": 1,
|
| 311 |
+
"tokens_per_sec": 14260.806253193065,
|
| 312 |
+
"step_time_ms": 397.53009557724,
|
| 313 |
+
"elapsed_sec": 1099.8754012584686,
|
| 314 |
+
"progress_pct": 34.716981132075475
|
| 315 |
+
},
|
| 316 |
+
{
|
| 317 |
+
"step": 2400,
|
| 318 |
+
"epoch": 0,
|
| 319 |
+
"loss": 2.7003683042526245,
|
| 320 |
+
"perplexity": 14.885213002672018,
|
| 321 |
+
"grad_norm": 0.9999998567906512,
|
| 322 |
+
"lr": 9.063444108761329e-05,
|
| 323 |
+
"cross_attn_scale": 1,
|
| 324 |
+
"tokens_per_sec": 14264.660562306666,
|
| 325 |
+
"step_time_ms": 397.0608687400818,
|
| 326 |
+
"elapsed_sec": 1147.6348791122437,
|
| 327 |
+
"progress_pct": 36.22641509433962
|
| 328 |
+
},
|
| 329 |
+
{
|
| 330 |
+
"step": 2500,
|
| 331 |
+
"epoch": 0,
|
| 332 |
+
"loss": 2.6987667942047118,
|
| 333 |
+
"perplexity": 14.861393263349989,
|
| 334 |
+
"grad_norm": 0.999999836542777,
|
| 335 |
+
"lr": 9.441087613293051e-05,
|
| 336 |
+
"cross_attn_scale": 1,
|
| 337 |
+
"tokens_per_sec": 14272.673599238884,
|
| 338 |
+
"step_time_ms": 392.08106756210327,
|
| 339 |
+
"elapsed_sec": 1194.8396270275116,
|
| 340 |
+
"progress_pct": 37.735849056603776
|
| 341 |
+
},
|
| 342 |
+
{
|
| 343 |
+
"step": 2600,
|
| 344 |
+
"epoch": 0,
|
| 345 |
+
"loss": 2.679299759864807,
|
| 346 |
+
"perplexity": 14.574883803144939,
|
| 347 |
+
"grad_norm": 0.9999998161421668,
|
| 348 |
+
"lr": 9.818731117824775e-05,
|
| 349 |
+
"cross_attn_scale": 1,
|
| 350 |
+
"tokens_per_sec": 14268.220487957708,
|
| 351 |
+
"step_time_ms": 400.62639236450195,
|
| 352 |
+
"elapsed_sec": 1243.1093292236328,
|
| 353 |
+
"progress_pct": 39.24528301886792
|
| 354 |
+
},
|
| 355 |
+
{
|
| 356 |
+
"step": 2700,
|
| 357 |
+
"epoch": 0,
|
| 358 |
+
"loss": 2.6831858372688293,
|
| 359 |
+
"perplexity": 14.631633124469138,
|
| 360 |
+
"grad_norm": 0.9999997879513947,
|
| 361 |
+
"lr": 9.999883900503343e-05,
|
| 362 |
+
"cross_attn_scale": 1,
|
| 363 |
+
"tokens_per_sec": 14271.451176654387,
|
| 364 |
+
"step_time_ms": 396.198148727417,
|
| 365 |
+
"elapsed_sec": 1290.832359790802,
|
| 366 |
+
"progress_pct": 40.75471698113208
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"step": 2800,
|
| 370 |
+
"epoch": 0,
|
| 371 |
+
"loss": 2.6590483808517456,
|
| 372 |
+
"perplexity": 14.28269094737908,
|
| 373 |
+
"grad_norm": 0.9999997752687187,
|
| 374 |
+
"lr": 9.999008031189225e-05,
|
| 375 |
+
"cross_attn_scale": 1,
|
| 376 |
+
"tokens_per_sec": 14273.344204546162,
|
| 377 |
+
"step_time_ms": 393.01880836486816,
|
| 378 |
+
"elapsed_sec": 1338.1313955783844,
|
| 379 |
+
"progress_pct": 42.264150943396224
|
| 380 |
+
},
|
| 381 |
+
{
|
| 382 |
+
"step": 2900,
|
| 383 |
+
"epoch": 0,
|
| 384 |
+
"loss": 2.640705282688141,
|
| 385 |
+
"perplexity": 14.023090363712633,
|
| 386 |
+
"grad_norm": 0.999999769824181,
|
| 387 |
+
"lr": 9.997273618929405e-05,
|
| 388 |
+
"cross_attn_scale": 1,
|
| 389 |
+
"tokens_per_sec": 14276.261923323767,
|
| 390 |
+
"step_time_ms": 395.9458804130554,
|
| 391 |
+
"elapsed_sec": 1385.7869172096252,
|
| 392 |
+
"progress_pct": 43.77358490566038
|
| 393 |
+
},
|
| 394 |
+
{
|
| 395 |
+
"step": 3000,
|
| 396 |
+
"epoch": 0,
|
| 397 |
+
"loss": 2.6228610467910767,
|
| 398 |
+
"perplexity": 13.775078404108285,
|
| 399 |
+
"grad_norm": 0.9999997660929014,
|
| 400 |
+
"lr": 9.99468096460588e-05,
|
| 401 |
+
"cross_attn_scale": 1,
|
| 402 |
+
"tokens_per_sec": 14278.078970730441,
|
| 403 |
+
"step_time_ms": 395.1349449157715,
|
| 404 |
+
"elapsed_sec": 1433.3739182949066,
|
| 405 |
+
"progress_pct": 45.283018867924525
|
| 406 |
+
},
|
| 407 |
+
{
|
| 408 |
+
"step": 3100,
|
| 409 |
+
"epoch": 0,
|
| 410 |
+
"loss": 2.6258543491363526,
|
| 411 |
+
"perplexity": 13.816373151600999,
|
| 412 |
+
"grad_norm": 0.9999997773130892,
|
| 413 |
+
"lr": 9.991230517986574e-05,
|
| 414 |
+
"cross_attn_scale": 1,
|
| 415 |
+
"tokens_per_sec": 14279.949787692021,
|
| 416 |
+
"step_time_ms": 398.6564254760742,
|
| 417 |
+
"elapsed_sec": 1481.378668308258,
|
| 418 |
+
"progress_pct": 46.79245283018868
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"step": 3200,
|
| 422 |
+
"epoch": 0,
|
| 423 |
+
"loss": 2.6306656908988955,
|
| 424 |
+
"perplexity": 13.883008619210889,
|
| 425 |
+
"grad_norm": 0.9999997908261178,
|
| 426 |
+
"lr": 9.986922877647319e-05,
|
| 427 |
+
"cross_attn_scale": 1,
|
| 428 |
+
"tokens_per_sec": 14284.573973594528,
|
| 429 |
+
"step_time_ms": 397.2297763824463,
|
| 430 |
+
"elapsed_sec": 1529.228875875473,
|
| 431 |
+
"progress_pct": 48.301886792452834
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"step": 3300,
|
| 435 |
+
"epoch": 0,
|
| 436 |
+
"loss": 2.631927456855774,
|
| 437 |
+
"perplexity": 13.900536782760202,
|
| 438 |
+
"grad_norm": 0.9999998078319935,
|
| 439 |
+
"lr": 9.981758790868019e-05,
|
| 440 |
+
"cross_attn_scale": 1,
|
| 441 |
+
"tokens_per_sec": 14286.864134417545,
|
| 442 |
+
"step_time_ms": 395.89094400405884,
|
| 443 |
+
"elapsed_sec": 1576.9046158790588,
|
| 444 |
+
"progress_pct": 49.81132075471698
|
| 445 |
+
},
|
| 446 |
+
{
|
| 447 |
+
"step": 3400,
|
| 448 |
+
"epoch": 0,
|
| 449 |
+
"loss": 2.612626371383667,
|
| 450 |
+
"perplexity": 13.634813952868482,
|
| 451 |
+
"grad_norm": 0.999999823446259,
|
| 452 |
+
"lr": 9.975739153503001e-05,
|
| 453 |
+
"cross_attn_scale": 1,
|
| 454 |
+
"tokens_per_sec": 14249.206863483647,
|
| 455 |
+
"step_time_ms": 395.3779458999634,
|
| 456 |
+
"elapsed_sec": 1629.021897315979,
|
| 457 |
+
"progress_pct": 51.320754716981135
|
| 458 |
+
},
|
| 459 |
+
{
|
| 460 |
+
"step": 3500,
|
| 461 |
+
"epoch": 0,
|
| 462 |
+
"loss": 2.613145899772644,
|
| 463 |
+
"perplexity": 13.641899466198712,
|
| 464 |
+
"grad_norm": 0.9999998209327766,
|
| 465 |
+
"lr": 9.968865009825616e-05,
|
| 466 |
+
"cross_attn_scale": 1,
|
| 467 |
+
"tokens_per_sec": 14250.43626164504,
|
| 468 |
+
"step_time_ms": 397.92346715927124,
|
| 469 |
+
"elapsed_sec": 1676.8955392837524,
|
| 470 |
+
"progress_pct": 52.83018867924528
|
| 471 |
+
},
|
| 472 |
+
{
|
| 473 |
+
"step": 3600,
|
| 474 |
+
"epoch": 0,
|
| 475 |
+
"loss": 2.6072049522399903,
|
| 476 |
+
"perplexity": 13.561093925630912,
|
| 477 |
+
"grad_norm": 0.9999998081712715,
|
| 478 |
+
"lr": 9.961137552347076e-05,
|
| 479 |
+
"cross_attn_scale": 1,
|
| 480 |
+
"tokens_per_sec": 14254.882426479446,
|
| 481 |
+
"step_time_ms": 395.50578355789185,
|
| 482 |
+
"elapsed_sec": 1724.531726360321,
|
| 483 |
+
"progress_pct": 54.339622641509436
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"step": 3700,
|
| 487 |
+
"epoch": 0,
|
| 488 |
+
"loss": 2.608313500881195,
|
| 489 |
+
"perplexity": 13.576135393434422,
|
| 490 |
+
"grad_norm": 0.9999998063108952,
|
| 491 |
+
"lr": 9.95255812160958e-05,
|
| 492 |
+
"cross_attn_scale": 1,
|
| 493 |
+
"tokens_per_sec": 14254.472177341095,
|
| 494 |
+
"step_time_ms": 393.57229948043823,
|
| 495 |
+
"elapsed_sec": 1771.8527691364288,
|
| 496 |
+
"progress_pct": 55.84905660377358
|
| 497 |
+
},
|
| 498 |
+
{
|
| 499 |
+
"step": 3800,
|
| 500 |
+
"epoch": 0,
|
| 501 |
+
"loss": 2.6075013875961304,
|
| 502 |
+
"perplexity": 13.565114509230531,
|
| 503 |
+
"grad_norm": 0.999999798229707,
|
| 504 |
+
"lr": 9.943128205953757e-05,
|
| 505 |
+
"cross_attn_scale": 1,
|
| 506 |
+
"tokens_per_sec": 14255.298218348358,
|
| 507 |
+
"step_time_ms": 395.9108328819275,
|
| 508 |
+
"elapsed_sec": 1819.4854714870453,
|
| 509 |
+
"progress_pct": 57.35849056603774
|
| 510 |
+
},
|
| 511 |
+
{
|
| 512 |
+
"step": 3900,
|
| 513 |
+
"epoch": 0,
|
| 514 |
+
"loss": 2.5958344030380247,
|
| 515 |
+
"perplexity": 13.40777017944122,
|
| 516 |
+
"grad_norm": 0.9999998088575961,
|
| 517 |
+
"lr": 9.932849441260476e-05,
|
| 518 |
+
"cross_attn_scale": 1,
|
| 519 |
+
"tokens_per_sec": 14254.43387688602,
|
| 520 |
+
"step_time_ms": 397.1994471549988,
|
| 521 |
+
"elapsed_sec": 1867.6622467041016,
|
| 522 |
+
"progress_pct": 58.867924528301884
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"step": 4000,
|
| 526 |
+
"epoch": 0,
|
| 527 |
+
"loss": 2.599056921005249,
|
| 528 |
+
"perplexity": 13.4510466518991,
|
| 529 |
+
"grad_norm": 0.9999998203838558,
|
| 530 |
+
"lr": 9.921723610667061e-05,
|
| 531 |
+
"cross_attn_scale": 1,
|
| 532 |
+
"tokens_per_sec": 14254.509174407756,
|
| 533 |
+
"step_time_ms": 397.4755930900574,
|
| 534 |
+
"elapsed_sec": 1915.4945755004883,
|
| 535 |
+
"progress_pct": 60.37735849056604
|
| 536 |
+
},
|
| 537 |
+
{
|
| 538 |
+
"step": 4100,
|
| 539 |
+
"epoch": 0,
|
| 540 |
+
"loss": 2.6074876379966736,
|
| 541 |
+
"perplexity": 13.564927995621689,
|
| 542 |
+
"grad_norm": 0.9999998161921769,
|
| 543 |
+
"lr": 9.909752644257946e-05,
|
| 544 |
+
"cross_attn_scale": 1,
|
| 545 |
+
"tokens_per_sec": 14255.783417488716,
|
| 546 |
+
"step_time_ms": 392.7994465827942,
|
| 547 |
+
"elapsed_sec": 1962.7633347511292,
|
| 548 |
+
"progress_pct": 61.886792452830186
|
| 549 |
+
},
|
| 550 |
+
{
|
| 551 |
+
"step": 4200,
|
| 552 |
+
"epoch": 0,
|
| 553 |
+
"loss": 2.5968322062492373,
|
| 554 |
+
"perplexity": 13.421155172265358,
|
| 555 |
+
"grad_norm": 0.9999998175522805,
|
| 556 |
+
"lr": 9.896938618729855e-05,
|
| 557 |
+
"cross_attn_scale": 1,
|
| 558 |
+
"tokens_per_sec": 14258.306399981708,
|
| 559 |
+
"step_time_ms": 392.18066692352295,
|
| 560 |
+
"elapsed_sec": 2009.95512342453,
|
| 561 |
+
"progress_pct": 63.39622641509434
|
| 562 |
+
},
|
| 563 |
+
{
|
| 564 |
+
"step": 4300,
|
| 565 |
+
"epoch": 0,
|
| 566 |
+
"loss": 2.6084390473365784,
|
| 567 |
+
"perplexity": 13.577839936108278,
|
| 568 |
+
"grad_norm": 0.9999998215254254,
|
| 569 |
+
"lr": 9.883283757031538e-05,
|
| 570 |
+
"cross_attn_scale": 1,
|
| 571 |
+
"tokens_per_sec": 14261.791460845148,
|
| 572 |
+
"step_time_ms": 391.4139151573181,
|
| 573 |
+
"elapsed_sec": 2057.1161119937897,
|
| 574 |
+
"progress_pct": 64.90566037735849
|
| 575 |
+
},
|
| 576 |
+
{
|
| 577 |
+
"step": 4400,
|
| 578 |
+
"epoch": 0,
|
| 579 |
+
"loss": 2.583339796066284,
|
| 580 |
+
"perplexity": 13.24128759378564,
|
| 581 |
+
"grad_norm": 0.9999998188725712,
|
| 582 |
+
"lr": 9.868790427978143e-05,
|
| 583 |
+
"cross_attn_scale": 1,
|
| 584 |
+
"tokens_per_sec": 14263.497087724309,
|
| 585 |
+
"step_time_ms": 395.3731894493103,
|
| 586 |
+
"elapsed_sec": 2104.68616604805,
|
| 587 |
+
"progress_pct": 66.41509433962264
|
| 588 |
+
},
|
| 589 |
+
{
|
| 590 |
+
"step": 4500,
|
| 591 |
+
"epoch": 0,
|
| 592 |
+
"loss": 2.5934228324890136,
|
| 593 |
+
"perplexity": 13.375475352027253,
|
| 594 |
+
"grad_norm": 0.9999998173451519,
|
| 595 |
+
"lr": 9.853461145840274e-05,
|
| 596 |
+
"cross_attn_scale": 1,
|
| 597 |
+
"tokens_per_sec": 14265.037944890495,
|
| 598 |
+
"step_time_ms": 395.04913806915283,
|
| 599 |
+
"elapsed_sec": 2152.251477956772,
|
| 600 |
+
"progress_pct": 67.9245283018868
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"step": 4600,
|
| 604 |
+
"epoch": 0,
|
| 605 |
+
"loss": 2.599502420425415,
|
| 606 |
+
"perplexity": 13.457040420394211,
|
| 607 |
+
"grad_norm": 0.9999998118705633,
|
| 608 |
+
"lr": 9.837298569907825e-05,
|
| 609 |
+
"cross_attn_scale": 1,
|
| 610 |
+
"tokens_per_sec": 14267.814696965861,
|
| 611 |
+
"step_time_ms": 389.1418123245239,
|
| 612 |
+
"elapsed_sec": 2199.072714805603,
|
| 613 |
+
"progress_pct": 69.43396226415095
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"step": 4700,
|
| 617 |
+
"epoch": 0,
|
| 618 |
+
"loss": 2.598004117012024,
|
| 619 |
+
"perplexity": 13.436892788200378,
|
| 620 |
+
"grad_norm": 0.9999998030281144,
|
| 621 |
+
"lr": 9.820305504028653e-05,
|
| 622 |
+
"cross_attn_scale": 1,
|
| 623 |
+
"tokens_per_sec": 14273.25087695287,
|
| 624 |
+
"step_time_ms": 386.8022680282593,
|
| 625 |
+
"elapsed_sec": 2245.627872467041,
|
| 626 |
+
"progress_pct": 70.94339622641509
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"step": 4800,
|
| 630 |
+
"epoch": 0,
|
| 631 |
+
"loss": 2.5913213515281677,
|
| 632 |
+
"perplexity": 13.3473965590906,
|
| 633 |
+
"grad_norm": 0.9999998061302938,
|
| 634 |
+
"lr": 9.802484896122164e-05,
|
| 635 |
+
"cross_attn_scale": 1,
|
| 636 |
+
"tokens_per_sec": 14279.499985340064,
|
| 637 |
+
"step_time_ms": 390.2846693992615,
|
| 638 |
+
"elapsed_sec": 2292.609687566757,
|
| 639 |
+
"progress_pct": 72.45283018867924
|
| 640 |
+
},
|
| 641 |
+
{
|
| 642 |
+
"step": 4900,
|
| 643 |
+
"epoch": 0,
|
| 644 |
+
"loss": 2.5952907371520997,
|
| 645 |
+
"perplexity": 13.400482813314012,
|
| 646 |
+
"grad_norm": 0.9999998140407943,
|
| 647 |
+
"lr": 9.783839837667931e-05,
|
| 648 |
+
"cross_attn_scale": 1,
|
| 649 |
+
"tokens_per_sec": 14285.269544087463,
|
| 650 |
+
"step_time_ms": 389.3049716949463,
|
| 651 |
+
"elapsed_sec": 2339.454911708832,
|
| 652 |
+
"progress_pct": 73.9622641509434
|
| 653 |
+
}
|
| 654 |
+
],
|
| 655 |
+
"checkpoint": {
|
| 656 |
+
"pct": 75,
|
| 657 |
+
"step": 4968,
|
| 658 |
+
"final_loss": 2.591787926469864,
|
| 659 |
+
"total_tokens": 33883181,
|
| 660 |
+
"total_time_sec": 2371.5936682224274
|
| 661 |
+
}
|
| 662 |
+
}
|