初始化项目，由ModelHub XC社区提供模型

Model: InosLihka/rhythm-env-meta-trained-iter1 Source: Original Platform
2026-05-16 23:06:03 +08:00
commit 02f436bc80
36 changed files with 319837 additions and 0 deletions
--- a/training_config.json
+++ b/training_config.json
@@ -0,0 +1,15 @@
+{
+  "model_name": "unsloth/Qwen2.5-3B-Instruct",
+  "max_steps": 200,
+  "num_episodes": 80,
+  "max_samples": 800,
+  "num_generations": 4,
+  "learning_rate": 5e-05,
+  "beta": 0.04,
+  "lora_rank": 8,
+  "hint_fraction": 0.15,
+  "profile_mode": "continuous",
+  "output_dir": "/tmp/rhythm_env/outputs/rhythmenv_meta_trained",
+  "use_simple_reward": false,
+  "report_to": "none"
+}