app.py

import gradio as gr
import onnxruntime as ort
import numpy as np
from transformers import AutoTokenizer

# Load tokenizer (Qwen tokenizer)
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B")

# Load ONNX model
session = ort.InferenceSession("/models/model.onnx")

def chat_fn(user_input, history):
    if not user_input.strip():
        return history

    # Tokenize input
    inputs = tokenizer(user_input, return_tensors="np", padding=True)
    ort_inputs = {session.get_inputs()[0].name: inputs["input_ids"].astype(np.int64)}

    # Run inference
    output = session.run(None, ort_inputs)[0]

    # Decode model output
    text = tokenizer.decode(output[0], skip_special_tokens=True)
    history.append(("🧑‍💻 You: " + user_input, "🤖 Sam (Qwen3): " + text))
    return history

demo = gr.ChatInterface(fn=chat_fn, title="💬 Qwen3-0.6B-ONNX Demo",
                        description="Running ONNX model on a prebuilt Docker Space (SmilyAI Style!)")

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