| EVALUATION SUMMARY | |
| ================================================================================================== | |
| Checkpoint Exact Match Token Acc BLEU PERPLEXITY ms/sample samp/s | |
| -------------------------------------------------------------------------------------------------- | |
| initial_checkpoint 0.00% 27.11% 1.00 692.82 34.5 29.02 | |
| checkpoint_step_3000 28.88% 31.43% 15.87 335.37 15.4 65.14 | |
| checkpoint_step_6000 28.89% 32.19% 16.03 309.61 15.4 64.93 | |
| checkpoint_step_9000 29.97% 31.97% 16.44 262.67 15.2 65.87 | |
| checkpoint_step_9887 29.71% 32.21% 15.74 294.75 15.5 64.36 | |
| checkpoint_step_12000 29.68% 32.14% 16.41 370.60 16.0 62.63 | |
| checkpoint_step_15000 30.36% 32.98% 16.78 387.80 16.2 61.85 | |
| checkpoint_step_18000 30.59% 32.98% 16.78 360.23 15.7 63.60 | |
| checkpoint_step_19774 30.48% 33.01% 16.86 363.14 15.6 64.05 | |
| checkpoint_step_21000 30.49% 33.06% 16.81 375.95 15.7 63.75 | |
| checkpoint_step_24000 30.28% 33.04% 16.71 383.29 15.7 63.64 | |
| checkpoint_step_27000 30.30% 32.94% 16.55 386.61 15.7 63.71 | |
| checkpoint_step_29661 30.38% 32.95% 16.75 345.30 15.2 65.77 | |
| checkpoint_latest 30.36% 32.90% 16.63 418.24 16.0 62.55 | |
| model_best 30.61% 32.95% 16.87 320.34 15.1 66.40 | |
| model_final 30.42% 32.94% 16.78 345.30 15.1 66.28 | |