app.py

import gradio as gr
from huggingface_hub import hf_hub_download
from llama_cpp import Llama
import os

# 强制设置环境，让 Gradio 知道它在 Spaces 运行
os.environ["GRADIO_ANALYTICS_ENABLED"] = "False"

repo_id = "unsloth/gemma-4-E2B-it-GGUF" 
filename = "gemma-4-E2B-it-Q4_K_M.gguf"

print("正在加载模型...")
model_path = hf_hub_download(repo_id=repo_id, filename=filename)
llm = Llama(model_path=model_path, n_ctx=2048, n_threads=2, chat_format="gemma")

def chat_with_gemma(prompt, history):
    messages = []
    for user_msg, bot_msg in history:
        messages.append({"role": "user", "content": user_msg})
        messages.append({"role": "assistant", "content": bot_msg})
    messages.append({"role": "user", "content": prompt})
    
    stream = llm.create_chat_completion(messages=messages, max_tokens=512, stream=True)
    
    response = ""
    for chunk in stream:
        delta = chunk['choices'][0]['delta']
        if 'content' in delta:
            response += delta['content']
            yield response

# 使用最简单的 Block 结构，绕过复杂的 ChatInterface 模板带来的 Jinja2 报错
with gr.Blocks() as demo:
    gr.Markdown("# Gemma 4 E2B (Docker CPU版)")
    chatbot = gr.Chatbot()
    msg = gr.Textbox()
    
    def respond(message, chat_history):
        bot_message = ""
        # 这里为了简化，直接调用 generate，也可以使用 generator
        messages = [{"role": "user", "content": message}]
        stream = llm.create_chat_completion(messages=messages, stream=True)
        for chunk in stream:
            delta = chunk['choices'][0]['delta']
            if 'content' in delta:
                bot_message += delta['content']
                chat_history.append((message, bot_message))
                yield "", chat_history
                chat_history.pop()
                
    msg.submit(respond, [msg, chatbot], [msg, chatbot])

if __name__ == "__main__":
    # 彻底避开一切参数校验
    # 只需这一行，在 Hugging Face Space 上它会自动完成所有配置
    demo.launch(debug=True)