from transformers import GenerationConfig, LlamaForCausalLM, LlamaTokenizer,AutoTokenizer,AutoModelForCausalLM,MistralForCausalLM
import torch

model_id=Mistral-7B-Instruct-v0.4

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id,torch_dtype=torch.bfloat16,device_map="auto",)

chat_template="{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}"

def chat_format(conversation:list):
    system_prompt = "You are a helpful, respectful and honest assistant.Help humman as much as you can."
    
    id = tokenizer.apply_chat_template(conversation,chat_template=chat_template,tokenize=False)
    
    return id

user_chat=[{"role":"user","content":"你好，最近在干嘛呢"}]
text = chat_format(user_chat).rstrip("</s>")
def predict(content_prompt):
    inputs = tokenizer(content_prompt,return_tensors="pt",add_special_tokens=True)
    input_ids = inputs["input_ids"].to("cuda:0")
    # print(f"input length:{len(input_ids[0])}")
    with torch.no_grad():
        generation_output = model.generate(
                    input_ids=input_ids,
                    #generation_config=generation_config,
                    return_dict_in_generate=True,
                    output_scores=True,
                    max_new_tokens=2048,
                    top_p=0.9,
                    num_beams=1,
                    do_sample=True,
                    repetition_penalty=1.0,
                    eos_token_id=tokenizer.eos_token_id,
                    pad_token_id=tokenizer.pad_token_id,
                )
        s = generation_output.sequences[0]
        output = tokenizer.decode(s,skip_special_tokens=True)
        output1 = output.split("[/INST]")[-1].strip()
        # print(output1)
    return output1

predict(text)
output：你好！作为一个大型语言模型，我一直在学习和提高自己的能力。最近，我一直在努力学习新知识、改进算法，以便更好地回答用户的问题并提供帮助。同时，我也会定期接受人工智能专家的指导和评估，以确保我的表现不断提升。希望这些信息对你有所帮助！

vLLM server

shell

#llama2-chat-template.jinja file is chat-template above
model_path=Mistral-7B-Instruct-V0.4
python  -m vllm.entrypoints.openai.api_server --model=$model_path \
        --trust-remote-code --host 0.0.0.0  --port 7777 \
        --gpu-memory-utilization 0.8 \
        --max-model-len 8192 --chat-template llama2-chat-template.jinja \
        --tensor-parallel-size 1 --served-model-name chatbot

python

from openai import OpenAI
# Set OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:7777/v1"

client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)
call_args = {
        'temperature': 0.7,
        'top_p': 0.9,
        'top_k': 40,
        'max_tokens': 2048, # output-len
        'presence_penalty': 1.0,
        'frequency_penalty': 0.0,
        "repetition_penalty":1.0,
        "stop":["</s>"],
    }
chat_response = client.chat.completions.create(
    model="chatbot",
    messages=[
        {"role": "user", "content": "你好"},
    ],
    extra_body=call_args
)
print("Chat response:", chat_response)

Moses25/Mistral-7B-Instruct-V0.4

This model is trained from Mistral-7B-Instruct-V0.2 with 90% chinese dataset and 10% english dataset

vLLM server

Moses25/
Mistral-7B-Instruct-V0.4