From 54b0d3d0e88c7c2130857d425b516c62ca58e7d8 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Sun, 2 Feb 2025 23:36:22 -0500 Subject: [PATCH] passthrough dataset parser for dpo/grpo --- src/axolotl/core/trainers/grpo/trainer.py | 2 ++ src/axolotl/prompt_strategies/dpo/passthrough.py | 14 ++++++++++++++ 2 files changed, 16 insertions(+) create mode 100644 src/axolotl/prompt_strategies/dpo/passthrough.py diff --git a/src/axolotl/core/trainers/grpo/trainer.py b/src/axolotl/core/trainers/grpo/trainer.py index c3cb23f8c..d138d3ec6 100644 --- a/src/axolotl/core/trainers/grpo/trainer.py +++ b/src/axolotl/core/trainers/grpo/trainer.py @@ -10,3 +10,5 @@ class AxolotlGRPOTrainer(SchedulerMixin, GRPOTrainer): """ Extend the base GRPOTrainer for axolotl helpers """ + + _tag_names = ["trl", "grpo", "axolotl"] diff --git a/src/axolotl/prompt_strategies/dpo/passthrough.py b/src/axolotl/prompt_strategies/dpo/passthrough.py new file mode 100644 index 000000000..1fcb838db --- /dev/null +++ b/src/axolotl/prompt_strategies/dpo/passthrough.py @@ -0,0 +1,14 @@ +""" +DPO prompt strategies passthrough/zero-processing strategy +""" + + +def default( + cfg, dataset_idx=0, **kwargs +): # pylint: disable=possibly-unused-variable,unused-argument + def transform_fn( + sample, tokenizer=None + ): # pylint: disable=possibly-unused-variable,unused-argument + return sample + + return transform_fn