ChatGLM-6B/utils.py

import os
from typing import Dict, Tuple, Union, Optional

from torch.nn import Module
from transformers import AutoModel


def auto_configure_device_map(num_gpus: int) -> Dict[str, int]:
    # transformer.word_embeddings 占用1层
    # transformer.final_layernorm 和 lm_head 占用1层
    # transformer.layers 占用 28 层
    # 总共30层分配到num_gpus张卡上
    num_trans_layers = 28
    per_gpu_layers = 30 / num_gpus

    # bugfix: 在linux中调用torch.embedding传入的weight,input不在同一device上,导致RuntimeError
    # windows下 model.device 会被设置成 transformer.word_embeddings.device
    # linux下 model.device 会被设置成 lm_head.device
    # 在调用chat或者stream_chat时,input_ids会被放到model.device上
    # 如果transformer.word_embeddings.device和model.device不同,则会导致RuntimeError
    # 因此这里将transformer.word_embeddings,transformer.final_layernorm,lm_head都放到第一张卡上
    device_map = {'transformer.word_embeddings': 0,
                  'transformer.final_layernorm': 0, 'lm_head': 0}

    used = 2
    gpu_target = 0
    for i in range(num_trans_layers):
        if used >= per_gpu_layers:
            gpu_target += 1
            used = 0
        assert gpu_target < num_gpus
        device_map[f'transformer.layers.{i}'] = gpu_target
        used += 1

    return device_map


def load_model_on_gpus(checkpoint_path: Union[str, os.PathLike], num_gpus: int = 2,
                       device_map: Optional[Dict[str, int]] = None, **kwargs) -> Module:
    if num_gpus < 2 and device_map is None:
        model = AutoModel.from_pretrained(checkpoint_path, trust_remote_code=True, **kwargs).half().cuda()
    else:
        from accelerate import load_checkpoint_and_dispatch

        model = AutoModel.from_pretrained(checkpoint_path, trust_remote_code=True, **kwargs)
        model = model.eval()

        if device_map is None:
            device_map = auto_configure_device_map(num_gpus)

        model = load_checkpoint_and_dispatch(
            model, checkpoint_path, device_map=device_map, offload_folder="offload", offload_state_dict=True).half()

    return model