cfg

training/README.md

@@ -38,7 +38,7 @@ First, activate the virtual environment you prepared.
 Example GRPO training usage:
 
 ```bash
-python3 -u train_grpo.py --config-path configs/external_generalisation --config-name math_curriculum_qwen_7b $@ 2>&1 | tee verl_output.log
+python3 -u train_grpo.py --config-path configs/external_generalisation --config-name math_qwen_3b $@ 2>&1 | tee verl_output.log
 ```
 
 Then, having saved this as a bash script such as `train.sh`, run it:

training/configs/external_generalisation/math_qwen_3b.yaml

@@ -54,7 +54,7 @@ data:
 actor_rollout_ref:
   hybrid_engine: True
   model:
-    path: Qwen/Qwen2.5-7B-Instruct
+    path: Qwen/Qwen2.5-3B-Instruct
     external_lib: null
     override_config: { }
     enable_gradient_checkpointing: True