app.py

"""
Stack 2.9 - HuggingFace Space
Gradio 6.x compatible
"""
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

print("Loading model...")

MODEL_NAME = "Qwen/Qwen2.5-Coder-1.5B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.float32,
    device_map="cpu",
    trust_remote_code=True,
    low_cpu_mem_usage=True
)
print("Model loaded!")

def generate(prompt, max_tokens, temperature):
    messages = [{"role": "user", "content": prompt}]
    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer([text], return_tensors="pt")
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=int(max_tokens),
            temperature=float(temperature),
            do_sample=True,
            pad_token_id=tokenizer.pad_token_id or tokenizer.eos_token_id
        )
    
    response = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)
    return response.strip()

demo = gr.Interface(
    fn=generate,
    inputs=[
        gr.Textbox(label="Prompt", placeholder="Write a Python function to calculate fibonacci...", lines=4),
        gr.Number(label="Max tokens", value=256, minimum=64, maximum=512),
        gr.Number(label="Temperature", value=0.7, minimum=0.1, maximum=1.0),
    ],
    outputs=gr.Textbox(label="Response", lines=10),
    title="Stack 2.9 Code Assistant",
    description="Powered by Qwen2.5-Coder-1.5B",
)

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