Fix in learn to ask (#107)

Author: chenyushuo

tuner/learn_to_ask/README.md

@@ -289,6 +289,8 @@ Also, make sure to update the `model_path` in `tuner/learn_to_ask/main.py` to po
 
 > 🔗 Learn more about Tinker Backend: [Tinker Backend Documentation](https://agentscope-ai.github.io/Trinity-RFT/en/main/tutorial/example_tinker_backend.html)
 
+> In this provided example, training is configured for 4 epochs. When using Tinker, the total token consumption is approximately 112 million tokens, resulting in an estimated cost of approximately 18 USD.
+
 ### Launch Training
 ```bash
 python tuner/learn_to_ask/main.py

tuner/learn_to_ask/main.py

@@ -249,7 +249,6 @@ async def learn2ask_judge(
         temperature=1.0,
         tensor_parallel_size=1,
         inference_engine_num=4,
-        reasoning_parser=None,
     )
     aux_models = {
         AUXILIARY_MODEL_NAME: TunerModelConfig(
@@ -259,7 +258,6 @@ async def learn2ask_judge(
             temperature=0.7,
             tensor_parallel_size=2,
             inference_engine_num=1,
-            reasoning_parser=None,
         ),
     }
     algorithm = AlgorithmConfig(