app.py

import gradio as gr
import json

def chat_completions(request: gr.Request):
    data = request.json
    response = {
        "id": "chatcmpl-123",
        "object": "chat.completion",
        "created": 1677652288,
        "choices": [{
            "index": 0,
            "message": {
                "role": "assistant",
                "content": f"Placeholder response. Received: {data['messages'][-1]['content']}"
            },
            "finish_reason": "stop"
        }],
        "usage": {
            "prompt_tokens": 9,
            "completion_tokens": 12,
            "total_tokens": 21
        }
    }
    return json.dumps(response)

with gr.Blocks() as demo:
    gr.Markdown("# Chat Completions API")
    gr.Markdown("Send a POST request to /v1/chat/completions")

demo.queue()
demo.launch()

app = gr.mount_gradio_app(demo, "/v1/chat/completions", chat_completions)