Files
qwen3-8b-base-new-dpo-ultra…/all_results.json
ModelHub XC 66a4e63baa 初始化项目,由ModelHub XC社区提供模型
Model: W-61/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039
Source: Original Platform
2026-05-29 20:55:22 +08:00

27 lines
1021 B
JSON

{
"epoch": 0.9989528795811519,
"eval_KL/chosen_KL_mean": -55.269832611083984,
"eval_KL/mean": -83.84635925292969,
"eval_KL/rejected_KL_mean": -112.42288208007812,
"eval_KL/std": 79.61448669433594,
"eval_fcm_dpo/beta": 0.0027746616397053003,
"eval_logits/chosen": 1.3518378734588623,
"eval_logits/rejected": 1.4095185995101929,
"eval_logps/chosen": -335.6864929199219,
"eval_logps/ref_chosen": -280.41668701171875,
"eval_logps/ref_rejected": -264.6455383300781,
"eval_logps/rejected": -377.0683898925781,
"eval_loss": 0.6268572807312012,
"eval_margin_dpo/margin_mean": 57.15304946899414,
"eval_margin_dpo/margin_std": 93.23528289794922,
"eval_runtime": 88.7646,
"eval_samples": 2000,
"eval_samples_per_second": 22.531,
"eval_steps_per_second": 2.816,
"total_flos": 0.0,
"train_loss": 5.04975961889111,
"train_runtime": 7451.4996,
"train_samples": 61135,
"train_samples_per_second": 8.204,
"train_steps_per_second": 0.064
}