Gyaan Setu models
The on-device AI models used by Gyaan Setu, an offline learning app for school children in English, Telugu and Hindi.
The app downloads each model once, keeps it on the device, and runs it in the browser with transformers.js and ONNX Runtime Web, so it keeps working without the internet. The files are copies of the original models, converted or quantized to ONNX. Nothing was retrained.
| Folder | Used for | Original model | Licence |
|---|---|---|---|
SmolLM2-360M-Instruct/ |
Tutor answers (English) | HuggingFaceTB/SmolLM2-360M-Instruct | Apache 2.0 |
indictrans2/en-indic/ |
English → Telugu/Hindi translation | ai4bharat/indictrans2-en-indic-dist-200M | MIT |
indictrans2/indic-en/ |
Telugu/Hindi → English translation | ai4bharat/indictrans2-indic-en-dist-200M | MIT |
tts/mms-tts-eng/, tts/mms-tts-tel/, tts/mms-tts-hin/ |
Read-aloud voices | facebook/mms-tts | CC-BY-NC 4.0 (non-commercial) |
whisper/whisper-base.en/ |
English voice input | openai/whisper-base.en (ONNX by Xenova) | MIT |
Each folder's own README has the details. Files named *.part-aa,
*.part-ab, … are one file split into parts; the app joins them after
downloading.
The MMS voices may only be used for non-commercial purposes.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support