minor comment.

Commented out the 'fa2' line in model_args section.

Author: shamanez

examples/agentic_demo/agent_val_rock_swe_qwen35_2b.yaml

@@ -62,6 +62,7 @@ reward_pretrain: Qwen/Qwen3.5-2B
 
 actor_train:
   model_args:
+    # fa2 doesn't work for now
     flash_attn: sdpa
     attn_implementation: sdpa
     disable_gradient_checkpointing: false