Apppublic

eulogik/bharat-tiny-llm-v3-demo

sourceHugging Faceapache-2.0updated 1d agoView on Hugging Face

0likes

App README

[image]

Bharat-Tiny-LLM v3 — Chat Demo

A 1.7B-parameter bilingual (Hindi / Hinglish / English) LLM. This page is fully static — it calls the model through Hugging Face Inference Providers (Featherless AI) straight from your browser. No download, no WebGPU needed.

Setup: paste a free read-only HF token (huggingface.co/settings/tokens); it stays in your browser's localStorage.

—🇮🇳 Devanagari token injection for 36.8% Hindi compression
—📊 56% on GSM8K-Hindi (grade-school math in Hindi)
—⚡ 26.6% fewer bits/char than base Qwen3-1.7B on held-out Hindi

Model: eulogik/Bharat-Tiny-LLM-v3 (Apache 2.0) · Offline GGUFs in the same repo.