Safetensors
English
llama
alignment-handbook
trl
dpo
Generated from Trainer
Llama-3.1-8B-Magpie-Align-v0.2 / all_results.json
flydust's picture
End of training
f98f101 verified
raw
history blame contribute delete
780 Bytes
{
"epoch": 0.9992652461425422,
"eval_logits/chosen": -0.8059436678886414,
"eval_logits/rejected": -0.8133836388587952,
"eval_logps/chosen": -829.99755859375,
"eval_logps/rejected": -1107.798828125,
"eval_loss": 0.33278918266296387,
"eval_rewards/accuracies": 0.8899999856948853,
"eval_rewards/chosen": -4.859663486480713,
"eval_rewards/margins": 2.7170379161834717,
"eval_rewards/rejected": -7.576701641082764,
"eval_runtime": 303.6984,
"eval_samples": 2000,
"eval_samples_per_second": 6.585,
"eval_steps_per_second": 0.412,
"total_flos": 0.0,
"train_loss": 0.41019999388775796,
"train_runtime": 41758.0505,
"train_samples": 97988,
"train_samples_per_second": 2.347,
"train_steps_per_second": 0.018
}