csdc-atl/buffer-baichuan2-13B-rag-8bits

1import torch2from typing import Dict, List, Any3from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline4from transformers.generation.utils import GenerationConfig5 6# get dtype7dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] == 8 else torch.float168 9class EndpointHandler:10    def __init__(self, path=""):11        # load the model12        self.model = AutoModelForCausalLM.from_pretrained(path, device_map="auto", torch_dtype=dtype, trust_remote_code=True)13        self.model.generation_config = GenerationConfig.from_pretrained(path)14        self.tokenizer = AutoTokenizer.from_pretrained(path, use_fast=False, trust_remote_code=True)15 16    def __call__(self, data: Any) -> List[List[Dict[str, float]]]:17        inputs = data.pop("inputs", data)18        # ignoring parameters! Default to configs in generation_config.json.19        messages = [{"role": "user", "content": inputs}]20        response = self.model.chat(self.tokenizer, messages)21        if torch.backends.mps.is_available():22            torch.mps.empty_cache()23        return [{'generated_text': response}]24