app.py

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

BASE_MODEL = "Qwen/Qwen3-4B-Instruct-2507"  # ou le base exact utilisé dans ton notebook
LORA_REPO  = "XenocodeRCE/Socrate_4b"

tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    device_map="auto",
    torch_dtype=torch.float16,
    trust_remote_code=True,
)
model = PeftModel.from_pretrained(model, LORA_REPO)
model.eval()

def generate(prompt, max_new_tokens=256, temperature=0.7):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        out = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            do_sample=True,
            temperature=temperature,
        )
    return tokenizer.decode(out[0], skip_special_tokens=True)

demo = gr.Interface(
    fn=generate,
    inputs=[
        gr.Textbox(lines=6, label="Prompt"),
        gr.Slider(1, 1024, value=256, step=1, label="max_new_tokens"),
        gr.Slider(0.0, 2.0, value=0.7, step=0.05, label="temperature"),
    ],
    outputs=gr.Textbox(lines=10, label="Output"),
)

demo.launch()