app.py

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

# لود مدل و توکنایزر
model_name = "mistralai/Mistral-7B-Instruct-v0.3"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="cpu", torch_dtype=torch.float32)

# تابع چت برای رابط کاربری
def chat_fn(message, history):
    prompt = f"<s>[INST] {message} [/INST]"
    inputs = tokenizer(prompt, return_tensors="pt")
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=100,
            do_sample=True,
            temperature=0.7
        )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response

# تابع برای API
def api_predict(message):
    prompt = f"<s>[INST] {message} [/INST]"
    inputs = tokenizer(prompt, return_tensors="pt")
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=100,
            do_sample=True,
            temperature=0.7
        )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return {"response": response}

# رابط کاربری و API
with gr.Blocks(title="Mistral 7B Chatbot") as demo:
    gr.Markdown("# چت‌بات Mistral 7B")
    gr.Markdown("این چت‌بات روی Spaces رایگان اجرا می‌شه و API هم داره!")
    chatbot = gr.ChatInterface(fn=chat_fn)
    gr.Interface(fn=api_predict, inputs="text", outputs="json", title="API Endpoint")

demo.launch(server_name="0.0.0.0", server_port=7860)