configs.py

# coding=utf-8
# coding=utf-8
# Copyright 2023 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.
import dataclasses
from dataclasses import dataclass, field
from typing import Any, Dict, List, NewType, Optional, Tuple


@dataclass
class ModelArguments:
    """
    Arguments pertaining to which model/config/tokenizer we are going to fine-tune.
    """

    base_model_revision: Optional[str] = field(
        default=None,
        metadata={"help": ("The base model checkpoint for weights initialization with PEFT adatpers.")},
    )
    model_name_or_path: Optional[str] = field(
        default=None,
        metadata={
            "help": (
                "The model checkpoint for weights initialization. Don't set if you want to train a model from scratch."
            )
        },
    )
    model_revision: str = field(
        default="main",
        metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
    )
    model_code_revision: str = field(default=None, metadata={"help": "The branch of the IFT model"})
    torch_dtype: Optional[str] = field(
        default=None,
        metadata={
            "help": (
                "Override the default `torch.dtype` and load the model under this dtype. If `auto` is passed, the "
                "dtype will be automatically derived from the model's weights."
            ),
            "choices": ["auto", "bfloat16", "float16", "float32"],
        },
    )
    trust_remote_code: bool = field(default=False, metadata={"help": "Trust remote code when loading a model."})
    use_flash_attention_2: bool = field(
        default=False,
        metadata={
            "help": (
                "Whether to use flash attention 2. You must install this manually by running `pip install flash-attn --no-build-isolation`"
            )
        },
    )
    use_peft: bool = field(
        default=False,
        metadata={"help": ("Whether to use PEFT or not for training.")},
    )
    lora_r: Optional[int] = field(
        default=16,
        metadata={"help": ("LoRA R value.")},
    )
    lora_alpha: Optional[int] = field(
        default=32,
        metadata={"help": ("LoRA alpha.")},
    )
    lora_dropout: Optional[float] = field(
        default=0.05,
        metadata={"help": ("LoRA dropout.")},
    )
    lora_target_modules: Optional[List[str]] = field(
        default=None,
        metadata={"help": ("LoRA target modules.")},
    )
    lora_modules_to_save: Optional[List[str]] = field(
        default=None,
        metadata={"help": ("Model layers to unfreeze & train")},
    )
    load_in_8bit: bool = field(default=False, metadata={"help": "use 8 bit precision"})
    load_in_4bit: bool = field(default=False, metadata={"help": "use 4 bit precision"})

    bnb_4bit_quant_type: Optional[str] = field(
        default="nf4", metadata={"help": "precise the quantization type (fp4 or nf4)"}
    )
    use_bnb_nested_quant: bool = field(default=False, metadata={"help": "use nested quantization"})

    def __post_init__(self):
        if self.load_in_8bit and self.load_in_4bit:
            raise ValueError("You can't use 8 bit and 4 bit precision at the same time")


@dataclass
class DataArguments:
    """
    Arguments pertaining to what data we are going to input our model for training and eval.
    """

    chat_template: Optional[str] = field(default=None, metadata={"help": "The chat template to use."})
    dataset_mixer: Optional[Dict[str, float]] = field(
        default=None,
        metadata={"help": ("Datasets and their proportions to be used for training ift/rl.")},
    )
    dataset_splits: Optional[List[str]] = field(
        default_factory=lambda: ["train", "test"],
        metadata={"help": ("List of train test splits to use in the dataset")},
    )
    max_train_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "For debugging purposes or quicker training, truncate the number of training examples to this "
                "value if set."
            )
        },
    )
    max_eval_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "For debugging purposes or quicker training, truncate the number of evaluation examples to this "
                "value if set."
            )
        },
    )
    preprocessing_num_workers: Optional[int] = field(
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
    truncation_side: Optional[str] = field(
        default=None, metadata={"help": "Truncation side to use for the tokenizer."}
    )