Hugging Face's logo

phate334
/
Breeze-ASR-26-GGML

Automatic Speech Recognition
Min Nan Chinese
Chinese
whisper.cpp
whisper
taiwanese-hokkien
taigi
low-resource-language
arxiv:2603.19259
whisper-cpp
quantized
q8_0
q5_0
q4_0
q4_1
Model card Files
xet
 Community
2
Breeze-ASR-26-GGML / QUANTIZATION.md
phate334's picture
phate334
Upload folder using huggingface_hub
6a48547 verified
preview code
|
Raw History Blame Contribute Delete
1.4 kB

Quantization Record: Breeze-ASR-26-GGML

Source

  • Model: MediaTek-Research/Breeze-ASR-26
  • Revision: 7b992682e7f5ceedd0a41ebec240f01ba469d19e
  • Backend: whisper.cpp
  • Quantization: q8_0,q5_0,q4_0,q4_1

Command 

artifacts/tools/whisper.cpp/build/bin/whisper-quantize artifacts/tools/whispercpp-base/ggml-model.bin artifacts/models/Breeze-ASR-26-GGML/ggml-model-q8_0.bin q8_0 && artifacts/tools/whisper.cpp/build/bin/whisper-quantize artifacts/tools/whispercpp-base/ggml-model.bin artifacts/models/Breeze-ASR-26-GGML/ggml-model-q5_0.bin q5_0 && artifacts/tools/whisper.cpp/build/bin/whisper-quantize artifacts/tools/whispercpp-base/ggml-model.bin artifacts/models/Breeze-ASR-26-GGML/ggml-model-q4_0.bin q4_0 && artifacts/tools/whisper.cpp/build/bin/whisper-quantize artifacts/tools/whispercpp-base/ggml-model.bin artifacts/models/Breeze-ASR-26-GGML/ggml-model-q4_1.bin q4_1

Runtime Versions

  • architecture: arm64
  • ctranslate2_version: 4.7.1
  • huggingface_hub_version: 1.15.0
  • openai_version: 2.37.0
  • os: Darwin
  • os_release: 24.6.0
  • python_version: 3.13.0
  • transformers_version: 5.8.1
  • whispercpp_git_revision: 968eebe77225d25e57a3f981da7c696310f0e881

Platform Notes

  • Converted to GGML for whisper.cpp quantized artifact evaluation.
  • GGML convention keeps multiple quantized files for the same model in one repository.