Jacklanchantin/fix online dpo rm args (#1233)

* add new args for reward_handler.create

* add qwen25_3b

* increase max seq len

* revert yaml

Author: jacklanchantin

src/fairseq2/recipes/lm/_online_finetune/_online_dpo.py

@@ -456,10 +456,13 @@ def create(
         vllm_reward_model = vllm_actors.get(config.vllm_reward_model_actor_name, None)
         reward_registry = self._context.get_registry(VLLMOutputRewardHandler)
         reward_handler = reward_registry.get(config.reward.name)
+        reward_name = config.reward.name
         reward = reward_handler.create(
             reward_model=vllm_reward_model,
+            reward_name=reward_name,
             reward_config=config.reward.config,
             gangs=gangs,
+            context=self._context,
         )
 
         # TODO: decide converter as part of the model handler