# Notices & Attribution

## 1. Bodhan AI / AI4Bharat Indic-Transcribe-Flex
This model is derived from **Indic-Transcribe-Flex**, developed by **Bodhan AI** and **AI4Bharat**.
- Upstream Repository: [bodhan-ai/indic-transcribe-flex](https://huggingface.co/bodhan-ai/indic-transcribe-flex)
- Upstream Commit: `d1f27e693a2d0dd5735492908720003bdac81712`
- License: Indic Open Model License v1.0 (see [LICENSE](LICENSE))
- Required Attribution: **"Built with Indic-Transcribe-Flex from Bodhan AI / AI4Bharat."**

## 2. NVIDIA Canary 1B v2 Base Architecture
The underlying neural architecture is based on **Canary-1B-v2**, developed by NVIDIA Corporation.
- Architecture: FastConformer Encoder + Causal Transformer Decoder (~1.22B parameters)
- Upstream Model: [nvidia/canary-1b-v2](https://huggingface.co/nvidia/canary-1b-v2)
- Upstream License: [Creative Commons Attribution 4.0 International (CC-BY-4.0)](https://creativecommons.org/licenses/by/4.0/)
- Attribution: **"Canary-1B-v2 developed by NVIDIA Corporation (licensed under CC-BY-4.0)."**

## 3. Derivative Modifications
- **Transformation**: Dynamic Activation + Static Weight INT8 Quantization (W8A8 dynamic linear) applied across 530 linear projections.
- **Export Format**: Modular Microsoft ONNX Runtime graph (`encoder.onnx` + `decoder.onnx`) optimized for edge on-device mobile inference.
- **Target Distribution**: Client-side local inference within on-device mobile applications. No hosted multi-tenant API services are provided.
- **Developed by**: Community Contributor (2026).