src/llamafactory/model/model_utils/rope.py

# Copyright 2024 LMSYS and the LlamaFactory team.
# Copyright 2023 Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li
#
# This code is inspired by the LMSYS's FastChat library.
# https://github.com/lm-sys/FastChat/blob/v0.2.30/fastchat/train/train.py
#
# 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 math
from typing import TYPE_CHECKING

from ...extras.logging import get_logger


if TYPE_CHECKING:
    from transformers import PretrainedConfig

    from ...hparams import ModelArguments


logger = get_logger(__name__)


def configure_rope(config: "PretrainedConfig", model_args: "ModelArguments", is_trainable: bool) -> None:
    if model_args.rope_scaling is None:
        return

    if not hasattr(config, "rope_scaling"):
        logger.warning("Current model does not support RoPE scaling.")
        return

    if model_args.model_max_length is not None:
        if is_trainable and model_args.rope_scaling == "dynamic":
            logger.warning(
                "Dynamic NTK scaling may not work well with fine-tuning. "
                "See: https://github.com/huggingface/transformers/pull/24653"
            )

        current_max_length = getattr(config, "max_position_embeddings", None)
        if current_max_length and model_args.model_max_length > current_max_length:
            logger.info(
                "Enlarge max model length from {} to {}.".format(current_max_length, model_args.model_max_length)
            )
            setattr(config, "max_position_embeddings", model_args.model_max_length)
            scaling_factor = float(math.ceil(model_args.model_max_length / current_max_length))
        else:
            logger.warning("Input length is smaller than max length. Consider increase input length.")
            scaling_factor = 1.0
    else:
        scaling_factor = 2.0

    setattr(config, "rope_scaling", {"type": model_args.rope_scaling, "factor": scaling_factor})
    logger.info(
        "Using {} scaling strategy and setting scaling factor to {}".format(model_args.rope_scaling, scaling_factor)
    )