app.py

# app.py
import gradio as gr
import torch
from transformers import pipeline, AutoProcessor, AutoModelForVision2Seq

# -------------------
# Load Whisper (STT) from Hugging Face
# -------------------
stt_pipe = pipeline(
    "automatic-speech-recognition",
    model="openai/whisper-small",   # Replace with your uploaded Whisper model repo
    device=0 if torch.cuda.is_available() else -1
)

# -------------------
# Load ChatDOC model
# -------------------
chatdoc_model_id = "Muhammadidrees/RaiyaChatDoc"  # replace with your uploaded repo
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float16 if device == "cuda" else torch.float32

processor = AutoProcessor.from_pretrained(chatdoc_model_id, trust_remote_code=True)
chatdoc_model = AutoModelForVision2Seq.from_pretrained(
    chatdoc_model_id,
    torch_dtype=dtype
).to(device)

# -------------------
# Chat function
# -------------------
def chat_with_doc(audio, message, history=[]):
    transcript = ""
    if audio is not None:
        result = stt_pipe(audio)
        transcript = result["text"]

    user_msg = message or transcript
    if not user_msg.strip():
        return history, "No input detected."

    history.append([user_msg, None])

    # System prompt (simplified for demo)
    system_prompt = "You are a medical doctor interviewing a patient. Respond helpfully."
    dialogue = "\n".join([f"Patient: {u}\nDoctor: {b}" for u, b in history if u and b])
    prompt = f"{system_prompt}\n\nConversation:\n{dialogue}\nPatient: {user_msg}\nDoctor:"

    inputs = processor(text=prompt, images=None, return_tensors="pt").to(device)
    with torch.inference_mode():
        outputs = chatdoc_model.generate(
            **inputs,
            max_new_tokens=200,
            do_sample=True,
            temperature=0.7
        )

    input_len = inputs["input_ids"].shape[1]
    gen_tokens = outputs[:, input_len:]
    response = processor.batch_decode(gen_tokens, skip_special_tokens=True)[0].strip()

    history[-1][1] = response
    return history, response

# -------------------
# Gradio UI
# -------------------
with gr.Blocks(title="ChatDOC") as demo:
    gr.Markdown("# 🩺 ChatDOC + Whisper\nTalk or type your symptoms.")

    chatbot = gr.Chatbot(height=400)
    msg = gr.Textbox(placeholder="Type your symptoms...")
    mic = gr.Audio(sources=["microphone"], type="filepath", label="🎤 Speak your symptoms")

    clear_btn = gr.Button("Clear Chat")

    state = gr.State([])

    def respond(audio, text, history):
        return chat_with_doc(audio, text, history)

    msg.submit(respond, [mic, msg, state], [chatbot, msg, state])
    mic.change(respond, [mic, msg, state], [chatbot, msg, state])
    clear_btn.click(lambda: ([], "", []), None, [chatbot, msg, state])

if __name__ == "__main__":
    demo.launch()