seperately include max_completion_len

src/axolotl/core/trainers/grpo/__init__.py

@@ -43,6 +43,9 @@ class GRPOStrategy:
                 ] = cfg.grpo_vllm_gpu_memory_utilization
         if cfg.grpo_num_generations:
             grpo_args_kwargs["num_generations"] = cfg.grpo_num_generations
+        grpo_args_kwargs["max_completion_length"] = (
+            cfg.max_completion_length or cfg.sequence_len
+        )
         return grpo_args_kwargs
 
     @classmethod

src/axolotl/utils/config/models/input/v0_4_1/__init__.py

@@ -757,6 +757,12 @@ class AxolotlInputConfig(
         default=512,
         json_schema_extra={"description": "maximum prompt length for RL training"},
     )
+    max_completion_length: Optional[int] = Field(
+        default=None,
+        json_schema_extra={
+            "description": "Maximum length of the completion for RL training"
+        },
+    )
     sample_packing: Optional[bool] = None
     sample_packing_group_size: Optional[int] = 100_000
     sample_packing_bin_size: Optional[int] = 200