Attribution — please retain with the weights

The original model is ARK-ASR by AutoArk. The research authors are Yu Lin, Yiming Wang, Runyuan Cai, and Xiaodong Zeng.

Original checkpoint: Edge0/ARK-ASR-0.6B
Original training, architecture and evaluation code: AutoArk/open-audio-opd
Paper: Data-Efficient On-Policy Distillation for Automatic Speech Recognition
Pinned source revision: 45776b56d58cdfb2e2eb632f7e110f38684633e0
Original weights license: Apache-2.0, retained in LICENSE.

The GGUF conversion builds on harshav's existing ARK converter and runtime, published with ARK-ASR-3B-GGUF. The underlying native inference engine is the work of the transcribe.cpp authors. Their complete MIT license and copyright notice are retained in CONVERTER-LICENSE. The reused archive and its checksum are recorded in conversion.json.

maxffarrell's contribution is the 0.6B adaptation, quantization, validation and packaging, with Codex assistance. It is a community format conversion, without retraining or original-author endorsement. It does not claim authorship of the original model, research, training, or reused converter/runtime.

The v2 GGUF embeds these credits, license and source links in its metadata. Please retain this file, NOTICE and both license files when redistributing.

Original research citation

@misc{lin2026dataefficientopd,
  title={Data-Efficient On-Policy Distillation for Automatic Speech Recognition},
  author={Lin, Yu and Wang, Yiming and Cai, Runyuan and Zeng, Xiaodong},
  year={2026},
  eprint={2605.28139},
  archivePrefix={arXiv},
  primaryClass={cs.AI},
  url={https://arxiv.org/abs/2605.28139}
}