🎓 ATAR Prediction System (ML Ensemble)
Powered by Gradient Boosting + Random Forest + Ridge Regression
Features:
- 🚀 Train on ZeroGPU with automatic HF Model Repo upload
- 🔮 Predict ATAR from subject marks (auto-loads model from HF)
- ☁️ No persistent storage needed - models live in HF Model Repo
Train ML Ensemble & Upload to Hugging Face
1000 50000
Instructions:
- Set
HF_TOKENenvironment variable in Space settings (write access)- Go to Space Settings → Variables and secrets
- Add secret:
HF_TOKEN= your token from https://huggingface.co/settings/tokens
- (Optional) Specify version like
v1.0.0,v1.1.0, etc. or leave empty for auto timestamp - Click "Train & Upload to HF"
- Model will be uploaded to
victor-academy/atar-predictor-ensemble - Each training creates a new version - no overwrites!
Versioning:
models/latest/- Always the newest modelmodels/v1.0.0/- Specific version you can roll back tometadata.json- Tracks all versions with metrics
ZeroGPU:
- Training uses GPU for 120 seconds (free tier)
- Inference uses GPU for 5 seconds per request
- All model storage handled via HF Model Repo
Predict ATAR (JSON API)
Note:
- Model auto-loads
latestversion on first API call if not manually loaded - Manually load a specific version to test different models
- All versions are preserved in HF Model Repo
- Public repos: No token needed for downloads
- Private repos: Set
HF_TOKENenvironment variable in Space settings