2025 06 11 dwani 1 ga launch

dwani 1 - launch

Open Source Multimodal API for Indian Languages

Run the dwani.ai stack on your machine with docker

Use the API with python and nodejs library

Setup- dwani.ai Server wget dwani.ai/compose.yml docker compose -f compose.yml -p dwami-api up -d

-- Setup API Key and URL for your dwani.ai server pip install dwani

npm install dwani

--

System Requirement

Minimum GPU VRAM - 48 GB

Tested on GH200 GPU at Lambda Labs

Compatible with H100 / L40s / A100

Require read access permission for Google/Gemma-3 and AI4BHARAT/Indic-F5 Huggingface model repository

--

Open weight models 1. Text + Vision : Gemma3-4b-instruct 2. Text to Speech : Indic-F5 3. Translation : IndicTrans2 4. Speech to Text : IndicConformer Multilingual/ Nemo

Library 1. rolmocr for documents API 2. fastapi/uvicorn for API server 3. pytorch/transformer for model serving 4. onnx for model serving