optimum/habana/trl/trainer/dpo_config.py

# Copyright 2022 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from dataclasses import dataclass
from typing import Dict, Literal, Optional

from trl.trainer.dpo_config import FDivergenceType

from ... import GaudiTrainingArguments


@dataclass
class GaudiDPOConfig(GaudiTrainingArguments):
    r"""
    Initialize GaudiDPOConfig.
        Adapted from https://github.com/huggingface/trl/blob/v0.9.6/trl/trainer/dpo_config.py#L33
        - inherit from GaudiTrainingArguments
    """

    beta: float = 0.1
    label_smoothing: float = 0
    loss_type: Literal[
        "sigmoid", "hinge", "ipo", "bco_pair", "sppo_hard", "nca_pair", "robust", "aot", "aot_pair", "exo_pair"
    ] = "sigmoid"
    label_pad_token_id: int = -100
    padding_value: Optional[int] = None
    truncation_mode: str = "keep_end"
    max_length: Optional[int] = None
    max_prompt_length: Optional[int] = None
    max_target_length: Optional[int] = None
    is_encoder_decoder: Optional[bool] = None
    disable_dropout: bool = True
    generate_during_eval: bool = False
    precompute_ref_log_probs: bool = False
    dataset_num_proc: Optional[int] = None
    model_init_kwargs: Optional[Dict] = None
    ref_model_init_kwargs: Optional[Dict] = None
    model_adapter_name: Optional[str] = None
    ref_adapter_name: Optional[str] = None
    reference_free: bool = False
    force_use_ref_model: bool = False
    f_divergence_type: Optional[FDivergenceType] = FDivergenceType.REVERSE_KL
    f_alpha_divergence_coef: Optional[float] = 1.0
    sync_ref_model: bool = False
    ref_model_mixup_alpha: float = 0.9
    ref_model_sync_steps: int = 64
    rpo_alpha: Optional[float] = None

    def __post_init__(self):
        if self.loss_type == "kto_pair":
            raise ValueError("Support for kto_pair has been removed in DPOTrainer. Please use KTOTrainer.")
        return super().__post_init__()