W-61/llama-3-8b-base-r-dpo-ultrafeedback-4xh200-batch-128-20260426-105614
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-r-dpo-ultrafeedback-4xh200-batch-128-20260426-105614
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-05-06 10:32:46 +08:00
commit 05ce659608
19 changed files with 3507 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
eval_results.json Normal file
View File

@@ -0,0 +1,18 @@
{
"epoch": 0.9989528795811519,
"eval_logits/chosen": -0.8717418909072876,
"eval_logits/rejected": -0.8549114465713501,
"eval_logps/chosen": -415.0469055175781,
"eval_logps/ref_chosen": -288.6414794921875,
"eval_logps/ref_rejected": -265.96160888671875,
"eval_logps/rejected": -453.6796875,
"eval_loss": 0.5324965119361877,
"eval_r_dpo/chosen_len": 286.97601318359375,
"eval_r_dpo/length_delta": 40.88800048828125,
"eval_r_dpo/regularization_term": 0.0,
"eval_r_dpo/rejected_len": 246.08799743652344,
"eval_runtime": 77.9812,
"eval_samples": 2000,
"eval_samples_per_second": 25.647,
"eval_steps_per_second": 3.206
}