W-61/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-s_star-0.4-20260425-111846
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

初始化项目，由ModelHub XC社区提供模型

Browse Source 
Model: W-61/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-s_star-0.4-20260425-111846
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-05-26 14:22:32 +08:00
commit 7f970a698a
498 changed files with 15206 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
eval_results.json Normal file
View File

@@ -0,0 +1,17 @@
{
"epoch": 0.9989528795811519,
"eval_fcm_dpo/beta": 0.0036579067818820477,
"eval_logits/chosen": -0.8699643611907959,
"eval_logits/rejected": -0.854047954082489,
"eval_logps/chosen": -504.4405212402344,
"eval_logps/ref_chosen": -287.8268127441406,
"eval_logps/ref_rejected": -266.9300231933594,
"eval_logps/rejected": -574.5838623046875,
"eval_loss": 0.582348108291626,
"eval_margin_dpo/margin_mean": 91.04013061523438,
"eval_margin_dpo/margin_std": 147.3343505859375,
"eval_runtime": 78.8341,
"eval_samples": 2000,
"eval_samples_per_second": 25.37,
"eval_steps_per_second": 3.171
}