How spark1805/meta-pytorch-hackathon works
This is a demonstration RL environment for classifying sample emails, with agent-facing FastAPI endpoints and a Gradio experimentation dashboard. The workflow is: reset to sample an email, inspect its state, submit a label/action, and receive a reward; users can also request grading or compare a baseline. Source confirms both interfaces call the environment directly. The README documents deployment and capabilities; sampled code does not show a separate persistent store or external service integration.
Main entry points: main.py FastAPI endpoints · env.py Email environment · app.py Gradio dashboard · models.py Action and results
Components 12
Interaction surfaces
- FastAPI endpoints
api/main.py - Gradio dashboard
app.py - Server launcher
server/app.py
RL environment
- Email environment
app/env.py - Email examples
app/data.py
Task and scoring
- Action and results
app/models.py - Task catalog
api/main.py - Action grader
api/main.py - Baseline scorer
api/main.py - Task score rules
app/tasks.py
Other components
- RL agent
- Dashboard user
Connections 14
- RL agent → FastAPI endpoints: requests
- Dashboard user → Gradio dashboard: interacts
- Server launcher → FastAPI endpoints: runs
- FastAPI endpoints → Email environment: resets, steps, reads state
- Gradio dashboard → Email environment: resets, steps, reads state
- Email environment → Email examples: samples
- FastAPI endpoints → Action and results: uses
- Gradio dashboard → Action and results: constructs actions
- FastAPI endpoints → Task catalog: serves
- FastAPI endpoints → Action grader: grades
- Action grader → Email environment: reads state
- Gradio dashboard → Email environment: reads state for grading
- FastAPI endpoints → Baseline scorer: computes
- Baseline scorer → Email environment: resets and steps