app.py

import os
import sys
import datetime

from openai import OpenAI
import gradio as gr
from gradio.components.chatbot import ChatMessage, Message
from typing import (
    Any,
    Literal,
)

DEBUG_LOG = False or os.environ.get("DEBUG_LOG") == "True"

print(f"Gradio version: {gr.__version__}")

title = None  # "ServiceNow-AI Chat" # modelConfig.get('MODE_DISPLAY_NAME')
description = "Please use the community section on this space to provide feedback! <a href=\"https://huggingface.co/ServiceNow-AI/Apriel-Nemotron-15b-Thinker/discussions\">ServiceNow-AI/Apriel-Nemotron-Chat</a>"

chat_start_count = 0

model_config = {
    "MODEL_NAME": os.environ.get("MODEL_NAME"),
    "MODE_DISPLAY_NAME": os.environ.get("MODE_DISPLAY_NAME"),
    "MODEL_HF_URL": os.environ.get("MODEL_HF_URL"),
    "VLLM_API_URL": os.environ.get("VLLM_API_URL"),
    "AUTH_TOKEN": os.environ.get("AUTH_TOKEN")
}

# Initialize the OpenAI client with the vLLM API URL and token
client = OpenAI(
    api_key=model_config.get('AUTH_TOKEN'),
    base_url=model_config.get('VLLM_API_URL')
)


def log_message(message):
    if DEBUG_LOG is True:
        print(message)


# Gradio 5.0.1 had issues with checking the message formats.  5.29.0 does not!
def _check_format(messages: Any, type: Literal["messages", "tuples"] = "messages") -> None:
    if type == "messages":
        all_valid = all(
            isinstance(message, dict)
            and "role" in message
            and "content" in message
            or isinstance(message, ChatMessage | Message)
            for message in messages
        )
        if not all_valid:
            # Display which message is not valid
            for i, message in enumerate(messages):
                if not (isinstance(message, dict) and
                        "role" in message and
                        "content" in message) and not isinstance(message, ChatMessage | Message):
                    print(f"_check_format() --> Invalid message at index {i}: {message}\n", file=sys.stderr)
                    break

            raise Exception(
                "Data incompatible with messages format. Each message should be a dictionary with 'role' and 'content' keys or a ChatMessage object."
            )
        # else:
        #     print("_check_format() --> All messages are valid.")
    elif not all(
            isinstance(message, (tuple, list)) and len(message) == 2
            for message in messages
    ):
        raise Exception(
            "Data incompatible with tuples format. Each message should be a list of length 2."
        )


def chat_fn(message, history):
    log_message(f"{'-' * 80}\nchat_fn() --> Message: {message}")

    global chat_start_count
    chat_start_count = chat_start_count + 1
    print(
        f"{datetime.datetime.now()}: chat_start_count: {chat_start_count}, turns: {int(len(history if history else []) / 3)}")

    # Remove any assistant messages with metadata from history for multiple turns
    log_message(f"Original History: {history}")
    _check_format(history, "messages")
    history = [item for item in history if
               not (isinstance(item, dict) and
                    item.get("role") == "assistant" and
                    isinstance(item.get("metadata"), dict) and
                    item.get("metadata", {}).get("title") is not None)]
    log_message(f"Updated History: {history}")
    _check_format(history, "messages")

    history.append({"role": "user", "content": message})
    log_message(f"History with user message: {history}")
    _check_format(history, "messages")

    # Create the streaming response
    stream = client.chat.completions.create(
        model=model_config.get('MODEL_NAME'),
        messages=history,
        temperature=0.8,
        stream=True
    )

    history.append(gr.ChatMessage(
        role="assistant",
        content="Thinking...",
        metadata={"title": "🧠 Thought"}
    ))
    log_message(f"History added thinking: {history}")
    _check_format(history, "messages")

    output = ""
    completion_started = False
    for chunk in stream:
        # Extract the new content from the delta field
        content = getattr(chunk.choices[0].delta, "content", "")
        output += content

        parts = output.split("[BEGIN FINAL RESPONSE]")

        if len(parts) > 1:
            if parts[1].endswith("[END FINAL RESPONSE]"):
                parts[1] = parts[1].replace("[END FINAL RESPONSE]", "")
            if parts[1].endswith("[END FINAL RESPONSE]\n<|end|>"):
                parts[1] = parts[1].replace("[END FINAL RESPONSE]\n<|end|>", "")

        history[-1 if not completion_started else -2] = gr.ChatMessage(
            role="assistant",
            content=parts[0],
            metadata={"title": "🧠 Thought"}
        )
        if completion_started:
            history[-1] = gr.ChatMessage(
                role="assistant",
                content=parts[1]
            )
        elif len(parts) > 1 and not completion_started:
            completion_started = True
            history.append(gr.ChatMessage(
                role="assistant",
                content=parts[1]
            ))

        # only yield the most recent assistant messages
        messages_to_yield = history[-1:] if not completion_started else history[-2:]
        # _check_format(messages_to_yield, "messages")
        yield messages_to_yield

    log_message(f"Final History: {history}")
    _check_format(history, "messages")


# Add the model display name and Hugging Face URL to the description
# description = f"### Model: [{MODE_DISPLAY_NAME}]({MODEL_HF_URL})"

print(f"Running model {model_config.get('MODE_DISPLAY_NAME')} ({model_config.get('MODEL_NAME')})")

gr.ChatInterface(
    chat_fn,
    title=title,
    description=description,
    theme=gr.themes.Default(primary_hue="green"),
    type="messages",
).launch()