Unable to load `Quantized` models

[kishorekaruppusamy]

GGML_ASSERT: /private/var/folders/q9/ms4_8l4970q525dlsjp_b9n0x6l4j9/T/pip-install-jqga24o3/whisper-cpp-python_e0149178a136485cadb677b46e61e602/vendor/whisper.cpp/ggml.c:4288: wtype != GGML_TYPE_COUNT

Facing Above Exception When loading Quantized Whisper Model

Python Version : Python 3.9.6 whisper_cpp_python version : 0.2.0 Macbook M1 Pro Model used : https://huggingface.co/ggerganov/whisper.cpp/blob/main/ggml-tiny-q5_1.bin

Kindly help me with this ..

[magicse]

Because it use old whisper.cpp without support Quantized models. You need compile new whisper.cpp libs by youself

[slavanorm]

@magicse , what would happen if repo would point vendor folder to latest whisper.cpp version?

[magicse]

@slavanorm This is whisper_cpp.py for new model and new whisper.cpp whisper_cpp.zip

Also you need build dll (libggml.dll , libwhisper.dll) from new whisper_cpp an put it to whisper_cpp_python root path.