Apppublic

puarun/vedicastro

sourceHugging Facemitupdated 2mo agoView on Hugging Face

0likes

App README

VedicAstro

Public Gradio demo (Hugging Face ZeroGPU / free Gradio Spaces) for Vedic charts (D1 / D9 / D10), Vimshottari dasa, gochar, and Q&A via Gemini.

Astrology math runs on CPU. Gemini is called over the API (no local model / no GPU required).

Space secrets

In Settings → Secrets add:

—GEMINI_API_KEY — from Google AI Studio

Optional:

—GEMINI_MODEL (default gemini-3.1-flash-lite)
—LLM_PROVIDER=gemini

Hardware

Use ZeroGPU (or free Gradio CPU if available). Do not use the Docker SDK on the free tier.

Local run (Gradio)

bash

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements-local.txt
export LLM_PROVIDER=gemini
export GEMINI_API_KEY=your_key_here
python gradio_app.py

Local run (FastAPI + Ollama)

bash

pip install -r requirements-local.txt
export LLM_PROVIDER=ollama
export OLLAMA_MODEL=qwen3:4b
ollama serve
python run.py

Branches

Branch	Purpose
`main`	Public Gradio demo + Gemini
`local-ai`	Ollama-focused local use

Notes

—Sidereal Lahiri ayanamsa, whole-sign houses
—Rahu/Ketu use the mean lunar node (traditional Jyotish / DrikPanchang). True node can differ by up to ~1.5–1.9° and flip the sign/house near boundaries.
—Swiss Ephemeris defaults to Fagan/Bradley (~0.9° from Lahiri). The app re-applies Lahiri on every calculation so positions stay aligned with DrikPanchang (typically within ~0.01°).
—Swiss Ephemeris data files download into app/ephe/ on first use (HF git rejects these binaries)
—Deploy to the Space without git binaries: python scripts/deploy_hf.py (needs huggingface_hub + hf auth login)
—Set TZ offset explicitly when possible (e.g. 5.5 for IST)