Skip to main content

core.trainers.async_grpo.trainer

core.trainers.async_grpo.trainer

Axolotl’s async GRPO trainer on TRL’s AsyncGRPOTrainer (see the package doc).

Classes

Name Description
AxolotlAsyncGRPOTrainer TRL’s AsyncGRPOTrainer adapted to axolotl and extended (see the package doc).

AxolotlAsyncGRPOTrainer

core.trainers.async_grpo.trainer.AxolotlAsyncGRPOTrainer(
    model,
    reward_funcs=None,
    args=None,
    train_dataset=None,
    processing_class=None,
    callbacks=None,
    optimizers=(None, None),
    peft_config=None,
    *,
    eval_dataset=None,
    reward_processing_classes=None,
    rollout_func=None,
    data_collator=None,
    trainer_scorers=None,
    **kwargs,
)

TRL’s AsyncGRPOTrainer adapted to axolotl and extended (see the package doc).