Skip to content

Latest commit

 

History

History
265 lines (207 loc) · 6.17 KB

API_TESTING.md

File metadata and controls

265 lines (207 loc) · 6.17 KB

Paper2Poster API Testing

This document explains how to test the Paper2Poster API using the provided Python scripts.

Test Scripts

1. simple_api_test.py - Basic Usage

A minimal script for quick testing:

# Basic usage
python simple_api_test.py paper.pdf

# Example output:
🎓 Generating poster for: paper.pdf
✅ API is healthy
📤 Uploading PDF and starting generation...
✅ Job started: abc123-def456-789
👀 Monitoring job progress...
📊 Status: processing - Parsing PDF paper...
📊 Status: processing - Generating poster layout...
📊 Status: processing - Generating PowerPoint presentation...
🎉 Generation completed!
💾 Downloading PowerPoint file...
✅ Downloaded: poster_abc123-def456-789.pptx (2,458,123 bytes)

🎉 Success! Poster saved as: poster_abc123-def456-789.pptx

2. test_api.py - Advanced Testing

A comprehensive script with full control over parameters:

Basic Usage

python test_api.py --pdf paper.pdf

Advanced Usage with Custom Parameters

python test_api.py \
    --pdf paper.pdf \
    --model-t 4o \
    --model-v 4o \
    --width 48 \
    --height 36 \
    --output-dir ./my_posters \
    --poll-interval 5

Using Local Models (vLLM)

# First start vLLM services
docker-compose --profile vllm up -d

# Then generate with local models
python test_api.py \
    --pdf paper.pdf \
    --model-t vllm_qwen \
    --model-v vllm_qwen_vl

Utility Commands

# List all jobs
python test_api.py --list-jobs

# Monitor existing job
python test_api.py --job-id abc123-def456-789

# Download files for existing job
python test_api.py --download-only abc123-def456-789

Prerequisites

  1. Start the API service:

    docker-compose up -d

  2. Install Python dependencies:

    pip install requests

  3. Verify API is running:

    curl http://localhost:6025/health

Usage Examples

Example 1: Quick Test

# Download a sample paper (example)
wget https://arxiv.org/pdf/2301.07041.pdf -O sample_paper.pdf

# Generate poster
python simple_api_test.py sample_paper.pdf

Example 2: Custom Configuration

python test_api.py \
    --pdf research_paper.pdf \
    --model-t 4o \
    --model-v 4o \
    --width 36 \
    --height 24 \
    --output-dir ./conference_posters

Example 3: Using Different Models

# GPT-4o models
python test_api.py --pdf paper.pdf --model-t 4o --model-v 4o

# Mix of models
python test_api.py --pdf paper.pdf --model-t 4o-mini --model-v 4o

# Local vLLM models (requires vLLM services running)
python test_api.py --pdf paper.pdf --model-t vllm_qwen --model-v vllm_qwen_vl

Example 4: Ablation Studies

# Disable tree layout
python test_api.py --pdf paper.pdf --ablation-no-tree-layout

# Disable commenter
python test_api.py --pdf paper.pdf --ablation-no-commenter

# Disable examples
python test_api.py --pdf paper.pdf --ablation-no-example

Script Parameters

test_api.py Parameters

Parameter Description Default
--pdf Path to PDF paper file Required
--api-url API base URL http://localhost:6025
--model-t Text model name 4o
--model-v Vision model name 4o
--width Poster width in inches 48
--height Poster height in inches 36
--output-dir Output directory ./downloads
--poll-interval Status check interval (seconds) 10
--no-blank-detection Disable blank detection False
--ablation-* Various ablation study options False

Available Models

Text Models (--model-t)

  • 4o - GPT-4o
  • 4o-mini - GPT-4o Mini
  • vllm_qwen - Qwen2.5-7B-Instruct (local)
  • claude-3-5-sonnet - Claude 3.5 Sonnet

Vision Models (--model-v)

  • 4o - GPT-4o Vision
  • vllm_qwen_vl - Qwen2.5-VL-7B-Instruct (local)
  • claude-3-5-sonnet - Claude 3.5 Sonnet

Expected Output Files

After successful generation, you'll get:

  • PowerPoint file (.pptx) - Editable poster presentation
  • PNG image (.png) - Static poster image

Troubleshooting

Common Issues

  1. API Connection Error

    ❌ Cannot connect to API: Connection refused

    Solution: Make sure the API service is running:

    docker-compose up -d
curl http://localhost:6025/health

  2. PDF Upload Error

    ❌ PDF file not found: paper.pdf

    Solution: Check the file path and ensure the PDF exists:

    ls -la paper.pdf
file paper.pdf  # Should show "PDF document"

  3. Job Timeout or Failure

    ❌ Job failed: OpenAI API key not found

    Solution: Check your environment variables:

    # Make sure .env file exists with your API keys
cat .env | grep API_KEY

  4. Model Not Available

    ❌ Model 'vllm_qwen' not available

    Solution: Start the vLLM services:

    docker-compose --profile vllm up -d

Debug Mode

For more detailed output, you can modify the scripts to add debug information:

import requests
import logging

# Enable debug logging
logging.basicConfig(level=logging.DEBUG)
requests.packages.urllib3.disable_warnings()

Manual API Testing

You can also test the API manually using curl:

# Health check
curl http://localhost:6025/health

# Generate poster
curl -X POST "http://localhost:6025/generate-poster" \
  -F "pdf_file=@paper.pdf" \
  -F "model_name_t=4o" \
  -F "model_name_v=4o"

# Check job status (replace JOB_ID)
curl http://localhost:6025/jobs/JOB_ID

# Download result (replace JOB_ID)
curl -O -J "http://localhost:6025/download/JOB_ID?file_type=pptx"

Performance Notes

  • Generation time: Typically 2-10 minutes depending on paper complexity
  • File sizes: PowerPoint files are usually 1-5 MB
  • Concurrent jobs: The API supports multiple simultaneous jobs
  • Rate limits: Depends on your AI model provider's limits

Support

If you encounter issues:

  1. Check the API logs: docker-compose logs paper2poster
  2. Verify your environment configuration
  3. Ensure all required services are running
  4. Review the API documentation at http://localhost:6025/docs