FIX: max_length and max_prompt_length was not being sent to ORPOTrainer (#1584)

* FIX: TRL trainer preprocessing step was running in one process

* FIX: max_length and max_prompt_length was not being sent to ORPOTrainer

* FIX: Change ORPO max prompt length to 1/4 of max length, otherwise we get strange behaviour

* FIX: Removed change from a different PR

* FIX: Black fix

* explicitly set max prompt len for orpo config

---------

Co-authored-by: Ali Mosavian <[email protected]>
Co-authored-by: Wing Lian <[email protected]>

src/axolotl/core/trainer_builder.py

@@ -1526,6 +1526,9 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
         if self.cfg.rl == "orpo":
             training_args_cls = ORPOConfig
             training_args_kwargs["dataset_num_proc"] = self.cfg.dataset_processes
+            training_args_kwargs["max_length"] = self.cfg.sequence_len
+            if self.cfg.max_prompt_len:
+                training_args_kwargs["max_prompt_length"] = self.cfg.max_prompt_len
 
         training_args = training_args_cls(
             per_device_train_batch_size=self.cfg.micro_batch_size,

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

@@ -517,6 +517,9 @@ class AxolotlInputConfig(
 
     sequence_len: int = Field(default=512)
     min_sample_len: Optional[int] = None
+    max_prompt_len: int = Field(
+        default=512, metadata={"help": "maximum prompt length for RL training"}
+    )
     sample_packing: Optional[bool] = None
     eval_sample_packing: Optional[bool] = None
     pad_to_sequence_len: Optional[bool] = None