README.md

metadata

title: Sanskrit D3PM Paraphrase
emoji: 🕉️
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 5.23.3
app_file: app.py
pinned: false

Sanskrit D3PM Gradio Space

This Space runs Roman/IAST Sanskrit to Devanagari generation.

Model Source

Set these Space variables in Settings → Variables and secrets:

• HF_CHECKPOINT_REPO = <your-username>/sanskrit-d3pm
• HF_CHECKPOINT_FILE = best_model.pt
• HF_CHECKPOINT_LABEL = main-model (optional)
• HF_DEFAULT_MODEL_TYPE = d3pm_cross_attention or d3pm_encoder_decoder
• HF_DEFAULT_INCLUDE_NEG = true or false
• HF_DEFAULT_NUM_STEPS = checkpoint diffusion steps, for example 4, 8, 16

The app will download checkpoint from your model repo and load it at runtime. If the model repo contains model_settings.json, the Space will use it automatically and these variables become optional overrides.

Optional MLflow Tracking in Space

You can enable lightweight MLflow event logging for inference + task runs. Set these optional variables in Settings → Variables and secrets:

• MLFLOW_TRACKING_URI (example: file:/tmp/mlruns or your remote tracking server URI)
• MLFLOW_EXPERIMENT_NAME (example: hf-space-sanskrit-d3pm)

If not set, the Space runs normally without MLflow.

Local Dev

pip install -r requirements.txt
python app.py