API_ENDPOINT_VERIFICATION.md

API Endpoint Verification Guide

Overview

This document explains how to verify that the api_name="safe_gpu_chat_handler" endpoint is properly deployed and running.

Implementation Status

✅ Code Deployed: The API endpoint registration code is present in app.py:

• Line 845: api_name="safe_gpu_chat_handler" is set on the api_message.submit() event
• Line 849: Log message confirms registration
• Lines 1899-1900, 2002-2003: Handler functions are initialized and set globally
• Lines 2073-2097: Verification logging on app startup

Verification Methods

1. Check Application Logs

When the application starts, you should see these log messages:

============================================================
VERIFYING API ENDPOINTS
============================================================
✓ API is enabled in Gradio config: True
Registered API endpoints:
  - /api/safe_gpu_chat_handler (POST)
  - /api/new_session (POST)
  - /api/update_session_info (POST)
  ...
✓ API handler function initialized: True
  Handler type: function
============================================================
LAUNCHING GRADIO APP
============================================================
API endpoints available at: http://<host>:7860/api/

Additionally, during interface creation:

✓ API endpoint '/api/safe_gpu_chat_handler' registered with api_name
✓ GPU chat handler initialized and set as API handler

OR

✓ Non-GPU chat handler initialized and set as API handler

2. Use the Verification Script

Run the verification script to test the endpoint:

python verify_api_endpoint.py [base_url]

Example:

# Local testing
python verify_api_endpoint.py http://localhost:7860

# HuggingFace Spaces
python verify_api_endpoint.py https://jatinautonomouslabs-research-ai-assistant.hf.space

The script will:

• Check if Gradio API info is accessible
• Verify the endpoint is registered
• Test actual endpoint accessibility
• Return exit code 0 if successful

3. Manual HTTP Test

Test the endpoint directly using curl or HTTP client:

curl -X POST http://localhost:7860/api/safe_gpu_chat_handler \
  -H "Content-Type: application/json" \
  -d '{
    "data": [
      "test message",
      [],
      "Admin_J",
      "Session: test123 | User: Admin_J | Interactions: 0"
    ]
  }'

Expected responses:

• 200 OK: Endpoint is working correctly
• 422 Validation Error: Endpoint exists but validation failed (still confirms deployment)
• 404 Not Found: Endpoint not registered

4. Check Gradio API Documentation

Visit the Gradio API documentation page:

• Local: http://localhost:7860/docs or http://localhost:7860/api/docs
• Spaces: https://<your-space>.hf.space/docs

Look for /api/safe_gpu_chat_handler in the list of available endpoints.

5. Inspect Gradio Config

If you have access to the running instance, check:

import gradio as gr
# After demo is created
demo.config.api_open  # Should be True
demo.config.show_api  # Should be True

Code Location Reference

Component	File	Line(s)
API Endpoint Registration	app.py	838-849
Handler Initialization (GPU)	app.py	1899-1900
Handler Initialization (Non-GPU)	app.py	2002-2003
Startup Verification	app.py	2073-2097
Global Handler Variable	app.py	27-28

Troubleshooting

Issue: Endpoint not found (404)

Check:

1. Application logs show "API endpoint registered" message
2. show_api=True is set in demo.launch() (line 2108)
3. Handler function is initialized (check logs for handler type)

Solution:

• Ensure api_name parameter is correctly set on the submit event
• Verify the handler function is defined before the interface is created

Issue: Handler function is None

Check:

1. Logs show handler initialization message
2. _api_chat_handler_fn is set after handlers are defined

Solution:

• Verify GPU availability check runs correctly
• Check that handler functions are defined (lines 1779-1997)

Issue: Endpoint accessible but returns errors

This is expected - The endpoint is deployed correctly. The errors are validation or processing errors, not deployment issues.

Expected Behavior

✅ Deployed Successfully:

• Application logs show registration messages
• HTTP requests to /api/safe_gpu_chat_handler return 200 or 422 (not 404)
• Verification script passes
• Handler function is not None

❌ Not Deployed:

• 404 errors on endpoint access
• Logs don't show registration messages
• Handler function is None

Additional Notes

• The endpoint uses hidden Gradio components (visible=False) to avoid UI interference
• The handler automatically selects GPU or non-GPU handler based on availability
• Fallback to process_message ensures endpoint always works
• All API endpoints are registered with api_name parameters for REST access