lambda-llama-3-8b-dpo-test-orca / eval_results.json
tanliboy's picture
End of training
a22b788 verified
raw
history blame
585 Bytes
{
"epoch": 0.9947643979057592,
"eval_logits/chosen": -2.4346187114715576,
"eval_logits/rejected": -1.9611175060272217,
"eval_logps/chosen": -562.0942993164062,
"eval_logps/rejected": -970.0545654296875,
"eval_loss": 0.12353485077619553,
"eval_rewards/accuracies": 0.9642857313156128,
"eval_rewards/chosen": -2.802823305130005,
"eval_rewards/margins": 3.8823659420013428,
"eval_rewards/rejected": -6.685189247131348,
"eval_runtime": 28.3026,
"eval_samples": 643,
"eval_samples_per_second": 22.719,
"eval_steps_per_second": 0.742
}