Quantized from the original FP16 ggml model to Q4_K
Tool: whisper.cpp (whisper-quantize)
No additional training or fine-tuning
The model architecture and weights were not altered beyond quantization
Files
ggml-breeze-asr-26-q4_k.bin (888.9 MB)
Notes
Quantization may introduce differences in output compared to the original model, including reduced accuracy in certain scenarios.
Attribution
This is an unofficial quantized derivative and is not an official release from MediaTek Research.
This repository is distributed under the Apache License 2.0, consistent with the upstream models.
No NOTICE file was provided in the upstream repositories at the time of publishing.
Disclaimer
This repository redistributes a quantized derivative of upstream models.
The original model, including its training data, licensing, and intended use, is provided by the original authors. This repository does not modify or make claims about those aspects beyond quantization.
Users are responsible for reviewing the upstream repositories and ensuring compliance with applicable licenses and use cases.
