PyTorch
llama
alignment-handbook
Generated from Trainer
MambaInLlama_0_50 / eval_results.json
Junxiong Wang
add models
a981c20
raw
history blame
570 Bytes
{
"epoch": 1.0,
"eval_logits/chosen": -1.203917384147644,
"eval_logits/rejected": -1.191305160522461,
"eval_logps/chosen": -463.00738525390625,
"eval_logps/rejected": -795.7894897460938,
"eval_loss": 0.4001549482345581,
"eval_rewards/accuracies": 0.8553571701049805,
"eval_rewards/chosen": -2.2448642253875732,
"eval_rewards/margins": 3.252180576324463,
"eval_rewards/rejected": -5.497044563293457,
"eval_runtime": 203.3522,
"eval_samples": 4461,
"eval_samples_per_second": 21.937,
"eval_steps_per_second": 0.344
}