# Niagara Batch English GGUF

This directory contains audio.cpp GGUF conversions of Applied Brain Research's Niagara Batch English ASR models.

Upstream models:
- 19M: https://huggingface.co/abr-ai/niagara-19m-batch.en
- 38M: https://huggingface.co/abr-ai/niagara-38m-batch.en

Converted files:
- `niagara-19m-batch.en-f32.gguf`
- `niagara-38m-batch.en-f32.gguf`

Both GGUF files store model floating-point tensors as F32 and include integer constant tensors from the upstream PT2 package. They are not Q8 or BF16 packages.

These files are two configurations of the same Niagara ASR model family. They are kept in one GGUF package directory so users can choose the 19M or 38M variant without treating them as separate model families.

The GGUF files were converted from the upstream ABR Niagara PT2 CPU packages for use with audio.cpp. They are converted forms of the upstream model weights; users must comply with the upstream Applied Brain Research Open License.

License:
- Applied Brain Research Open License v1.1
- License URL: https://www.appliedbrainresearch.com/license
- See `LICENSE` in this directory.

Important: this is not an unrestricted commercial model license. The ABR license permits non-commercial research use and permits any purpose only for users whose annual gross revenue, including affiliates, is under 1 million USD. If that revenue threshold is exceeded, use is limited to non-commercial research.