app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "my2000cup/Gaia-Petro-LLM"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

def build_prompt(history, system_message, user_message):
    # 可以根据你的模型模板调整
    messages = []
    if system_message:
        messages.append({"role": "system", "content": system_message})
    for user, assistant in history:
        if user:
            messages.append({"role": "user", "content": user})
        if assistant:
            messages.append({"role": "assistant", "content": assistant})
    messages.append({"role": "user", "content": user_message})
    # 使用 tokenizer 的 chat 模板
    prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    return prompt

def respond(
    message,
    history,
    system_message,
    max_tokens,
    temperature,
    top_p
):
    prompt = build_prompt(history, system_message, message)
    model_inputs = tokenizer([prompt], return_tensors="pt").to(model.device)

    # 流式输出
    streamer = None
    try:
        from transformers import TextIteratorStreamer
        streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    except ImportError:
        streamer = None

    gen_kwargs = dict(
        **model_inputs,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )
    if streamer:
        gen_kwargs["streamer"] = streamer
        thread = torch.Thread(target=model.generate, kwargs=gen_kwargs)
        thread.start()
        response = ""
        for new_text in streamer:
            response += new_text
            yield response
        thread.join()
    else:
        output = model.generate(**gen_kwargs)
        response = tokenizer.decode(output[0][model_inputs['input_ids'].shape[1]:], skip_special_tokens=True)
        yield response

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are an oil & gas industry expert.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

if __name__ == "__main__":
    demo.launch()