W-61/llama-3-8b-base-epsilon-dpo-ultrafeedback-8xh200
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-epsilon-dpo-ultrafeedback-8xh200
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-04-24 07:16:03 +08:00
commit 2efe520893
20 changed files with 5570 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
eval_results.json Normal file
View File

@@ -0,0 +1,20 @@
{
"epoch": 0.9989528795811519,
"eval_kl/n_epsilon_steps": 0.31703630089759827,
"eval_kl/p_epsilon_steps": 0.6743951439857483,
"eval_logits/chosen": -0.8084373474121094,
"eval_logits/rejected": -0.7665925025939941,
"eval_logps/chosen": -588.654052734375,
"eval_logps/ref_chosen": -287.9388427734375,
"eval_logps/ref_rejected": -266.7934875488281,
"eval_logps/rejected": -683.635009765625,
"eval_loss": 0.621621310710907,
"eval_rewards/accuracies": 0.6955645084381104,
"eval_rewards/chosen": -0.5053801536560059,
"eval_rewards/margins": 0.19223107397556305,
"eval_rewards/rejected": -0.6976111531257629,
"eval_runtime": 50.6489,
"eval_samples": 2000,
"eval_samples_per_second": 39.488,
"eval_steps_per_second": 1.244
}