初始化项目，由ModelHub XC社区提供模型

Model: CriteriaPO/qwen2.5-3b-dpo-finegrained
Source: Original Platform

all_results.json

@@ -0,0 +1,22 @@
+{
+    "epoch": 1.0,
+    "eval_logits/chosen": -2.9375,
+    "eval_logits/rejected": -2.625,
+    "eval_logps/chosen": -772.0,
+    "eval_logps/rejected": -1168.0,
+    "eval_loss": 0.21195410192012787,
+    "eval_rewards/accuracies": 0.9136363863945007,
+    "eval_rewards/chosen": -6.0,
+    "eval_rewards/margins": 4.1875,
+    "eval_rewards/rejected": -10.1875,
+    "eval_runtime": 3472.8109,
+    "eval_samples": 99217,
+    "eval_samples_per_second": 28.364,
+    "eval_steps_per_second": 0.443,
+    "total_flos": 0.0,
+    "train_loss": 0.0,
+    "train_runtime": 39.4392,
+    "train_samples": 1773152,
+    "train_samples_per_second": 44959.109,
+    "train_steps_per_second": 351.249
+}