app.py

import gradio as gr
from transformers import pipeline
import pypandoc
import os
import tempfile

# ----------------------
# 1. Load the chat model
# ----------------------
# The Tencent Hunyuan model is large; this will download weights the first time
# and fall back to CPU if a GPU isn't available.
chat_pipe = pipeline(
    "text-generation",
    model="tencent/Hunyuan-A13B-Instruct",
    trust_remote_code=True,
    device_map="auto",            # uses GPU if possible
    max_new_tokens=512,
)

# ----------------------
# 2. Chat helper
# ----------------------

def chat_ai(history, user_message):
    """LLM chat wrapper: takes the chat history + new user msg, returns assistant reply."""

    # Re‑create a messages list compatible with the model’s chat format
    messages = [{"role": "system", "content": "You are a helpful assistant."}]
    for user, bot in history:
        messages.append({"role": "user", "content": user})
        messages.append({"role": "assistant", "content": bot})
    messages.append({"role": "user", "content": user_message})

    # Generate
    completion = chat_pipe(messages)[0]["generated_text"]

    # Some chat models return the entire conversation; grab only the last assistant line
    assistant_reply = completion.split("\n")[-1].strip()
    return assistant_reply

# ----------------------
# 3. Document converter
# ----------------------
# Powered by Pandoc via the pypandoc wrapper. Make sure pandoc is available in the
# environment (apt-get install pandoc on Linux, or add it to requirements.txt).

SUPPORTED_TARGETS = [
    "pdf", "docx", "html", "md", "txt", "rtf", "odt", "epub"
]


def convert_document(file_obj, target_format):
    """Convert the uploaded file to the selected target format and return the path."""
    if file_obj is None:
        raise gr.Error("Please upload a file.")

    if target_format not in SUPPORTED_TARGETS:
        raise gr.Error(f"Unsupported target format: {target_format}")

    filename = os.path.basename(file_obj.name)
    src_ext = os.path.splitext(filename)[1].lstrip(".")

    with tempfile.TemporaryDirectory() as tmpdir:
        # Save the uploaded file to a temp path that Pandoc can read
        src_path = os.path.join(tmpdir, f"input.{src_ext}")
        with open(src_path, "wb") as f:
            f.write(file_obj.read())

        # Build output path inside tmp dir
        output_path = os.path.join(tmpdir, f"converted.{target_format}")

        # Run Pandoc
        try:
            pypandoc.convert_file(src_path, to=target_format, outputfile=output_path)
        except RuntimeError as e:
            raise gr.Error(f"Conversion failed: {e}")

        # Return as a downloadable file for Gradio
        return output_path

# ----------------------
# 4. Gradio UI
# ----------------------

with gr.Blocks(title="LLM Chat + Universal Document Converter") as demo:
    gr.Markdown(
        "# 🌐 LLM Chat + 📄 Document Converter\n"
        "Chat with **Hunyuan-A13B** (Tencent) or convert documents between multiple formats using **Pandoc**."
    )

    with gr.Tabs():
        # ----- Tab 1: Chat -----
        with gr.TabItem("Chat"):
            chatbot = gr.Chatbot()
            user_msg = gr.Textbox(placeholder="Ask me anything...", label="Your message")
            send_btn = gr.Button("Send")
            clear_btn = gr.Button("Clear")

            def _send(history, msg):
                history = history or []
                reply = chat_ai(history, msg)
                history.append((msg, reply))
                return history, ""

            send_btn.click(_send, inputs=[chatbot, user_msg], outputs=[chatbot, user_msg])
            clear_btn.click(lambda: [], None, chatbot)

        # ----- Tab 2: Document Converter -----
        with gr.TabItem("Convert"):
            file_input = gr.File(label="Upload document")
            target = gr.Dropdown(SUPPORTED_TARGETS, label="Convert to", value="pdf")
            convert_btn = gr.Button("Convert")
            result_file = gr.File(label="Download converted file")

            convert_btn.click(convert_document, inputs=[file_input, target], outputs=result_file)

    demo.launch()