langchain-ChatGLM/models/chatglm_llm.py

import json
from langchain.llms.base import LLM
from typing import Optional, List
from langchain.llms.utils import enforce_stop_tokens
from transformers import AutoTokenizer, AutoModel, AutoConfig
import torch
from configs.model_config import LLM_DEVICE
from langchain.callbacks.base import CallbackManager
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
from typing import Dict, Tuple, Union, Optional

DEVICE = LLM_DEVICE
DEVICE_ID = "0" if torch.cuda.is_available() else None
CUDA_DEVICE = f"{DEVICE}:{DEVICE_ID}" if DEVICE_ID else DEVICE


def torch_gc():
    if torch.cuda.is_available():
        with torch.cuda.device(CUDA_DEVICE):
            torch.cuda.empty_cache()
            torch.cuda.ipc_collect()


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


class ChatGLM(LLM):
    max_token: int = 10000
    temperature: float = 0.01
    top_p = 0.9
    # history = []
    tokenizer: object = None
    model: object = None
    history_len: int = 10
    streaming: bool = True
    callback_manager = CallbackManager([StreamingStdOutCallbackHandler()])

    def __init__(self):
        super().__init__()

    @property
    def _llm_type(self) -> str:
        return "ChatGLM"

    def _call(self,
              prompt: str,
              history: List[List[str]] = [],
              stop: Optional[List[str]] = None) -> str:
        if self.streaming:
            history = history + [[None, ""]]
            for stream_resp, history in self.model.stream_chat(
                    self.tokenizer,
                    prompt,
                    history=history[-self.history_len:] if self.history_len > 0 else [],
                    max_length=self.max_token,
                    temperature=self.temperature,
            ):
                yield stream_resp, history

        else:
            response, _ = self.model.chat(
                self.tokenizer,
                prompt,
                history=history[-self.history_len:] if self.history_len > 0 else [],
                max_length=self.max_token,
                temperature=self.temperature,
            )
            torch_gc()
            if stop is not None:
                response = enforce_stop_tokens(response, stop)
            history = history + [[None, response]]
            return response, history

    # def chat(self,
    #          prompt: str) -> str:
    #     response, _ = self.model.chat(
    #         self.tokenizer,
    #         prompt,
    #         history=self.history[-self.history_len:] if self.history_len > 0 else [],
    #         max_length=self.max_token,
    #         temperature=self.temperature,
    #     )
    #     torch_gc()
    #     self.history = self.history + [[None, response]]
    #     return response

    def load_model(self,
                   model_name_or_path: str = "THUDM/chatglm-6b",
                   llm_device=LLM_DEVICE,
                   use_ptuning_v2=False,
                   device_map: Optional[Dict[str, int]] = None,
                   **kwargs):
        self.tokenizer = AutoTokenizer.from_pretrained(
            model_name_or_path,
            trust_remote_code=True
        )

        model_config = AutoConfig.from_pretrained(model_name_or_path, trust_remote_code=True)

        if use_ptuning_v2:
            try:
                prefix_encoder_file = open('ptuning-v2/config.json', 'r')
                prefix_encoder_config = json.loads(prefix_encoder_file.read())
                prefix_encoder_file.close()
                model_config.pre_seq_len = prefix_encoder_config['pre_seq_len']
                model_config.prefix_projection = prefix_encoder_config['prefix_projection']
            except Exception as e:
                print(e)
                print("加载PrefixEncoder config.json失败")

        if torch.cuda.is_available() and llm_device.lower().startswith("cuda"):
            # 根据当前设备GPU数量决定是否进行多卡部署
            num_gpus = torch.cuda.device_count()
            if num_gpus < 2 and device_map is None:
                self.model = (
                    AutoModel.from_pretrained(
                        model_name_or_path,
                        config=model_config,
                        trust_remote_code=True,
                        **kwargs)
                    .half()
                    .cuda()
                )
            else:
                from accelerate import dispatch_model

                model = (
                    AutoModel.from_pretrained(
                        model_name_or_path,
                        trust_remote_code=True,
                        config=model_config,
                        **kwargs)
                    .half())
                # 可传入device_map自定义每张卡的部署情况
                if device_map is None:
                    device_map = auto_configure_device_map(num_gpus)

                self.model = dispatch_model(model, device_map=device_map)
        else:
            self.model = (
                AutoModel.from_pretrained(
                    model_name_or_path,
                    config=model_config,
                    trust_remote_code=True,
                    **kwargs)
                .float()
                .to(llm_device)
            )

        if use_ptuning_v2:
            try:
                prefix_state_dict = torch.load('ptuning-v2/pytorch_model.bin')
                new_prefix_state_dict = {}
                for k, v in prefix_state_dict.items():
                    if k.startswith("transformer.prefix_encoder."):
                        new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
                self.model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)
                self.model.transformer.prefix_encoder.float()
            except Exception as e:
                print(e)
                print("加载PrefixEncoder模型参数失败")

        self.model = self.model.eval()