failed to compute log mel spectrogram

[chigkim]

Hi,

I'm using M3 Max, and I built with CoreML support. When I run transcribe, it throws an error: "failed to compute log mel spectrogram."
I'm including the log below.
I'd appreciate any help! Thanks so much!

>>> from pywhispercpp.model import Model
>>> model = Model('models/ggml-medium.bin', n_threads=6)
[2024-05-20 07:48:47,691] {model.py:221} INFO - Initializing the model ...
whisper_init_from_file_with_params_no_state: loading model from 'models/ggml-medium.bin'
whisper_init_with_params_no_state: use gpu    = 1
whisper_init_with_params_no_state: flash attn = 0
whisper_init_with_params_no_state: gpu_device = 0
whisper_init_with_params_no_state: dtw        = 0
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51865
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 1024
whisper_model_load: n_audio_head  = 16
whisper_model_load: n_audio_layer = 24
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 1024
whisper_model_load: n_text_head   = 16
whisper_model_load: n_text_layer  = 24
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 4 (medium)
whisper_model_load: adding 1608 extra tokens
whisper_model_load: n_langs       = 99
whisper_backend_init: using Metal backend
ggml_metal_init: allocating
ggml_metal_init: found device: Apple M3 Max
ggml_metal_init: picking default device: Apple M3 Max
ggml_metal_init: default.metallib not found, loading from source
ggml_metal_init: GGML_METAL_PATH_RESOURCES = nil
ggml_metal_init: error: could not use bundle path to find ggml-metal.metal, falling back to trying cwd
ggml_metal_init: loading 'ggml-metal.metal'
ggml_metal_init: error: Error Domain=MTLLibraryErrorDomain Code=3 "program_source:3:10: fatal error: 'ggml-common.h' file not found
#include "ggml-common.h"
         ^~~~~~~~~~~~~~~
" UserInfo={NSLocalizedDescription=program_source:3:10: fatal error: 'ggml-common.h' file not found
#include "ggml-common.h"
         ^~~~~~~~~~~~~~~
}
whisper_backend_init: ggml_backend_metal_init() failed
whisper_model_load:      CPU total size =  1533.14 MB
whisper_model_load: model size    = 1533.14 MB
whisper_init_state: kv self size  =  150.99 MB
whisper_init_state: kv cross size =  150.99 MB
whisper_init_state: kv pad  size  =    6.29 MB
whisper_init_state: loading Core ML model from 'models/ggml-medium-encoder.mlmodelc'
whisper_init_state: first run on a device may take a while ...
whisper_init_state: Core ML model loaded
whisper_init_state: compute buffer (conv)   =    8.81 MB
whisper_init_state: compute buffer (cross)  =    7.85 MB
whisper_init_state: compute buffer (decode) =  142.09 MB
>>> print(Model.system_info())
AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | METAL = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | CUDA = 0 | COREML = 1 | OPENVINO = 0
>>> segments = model.transcribe(file, speed_up=True)
[2024-05-20 07:49:11,958] {model.py:130} INFO - Transcribing ...
whisper_full_with_state: failed to compute log mel spectrogram
[2024-05-20 07:49:11,958] {model.py:133} INFO - Inference time: 0.000 s

[absadiki]

Hi @chigkim,

I think it's the same issue encountered in #35, please check this solution.

[chigkim]

Thanks for the response.
After following the suggestion from #35, it doesn't complain about "fatal error: 'ggml-common.h' file not found" during loading the model.
However, model.transcribe still throws the error: "failed to compute log mel spectrogram."

>>> from pywhispercpp.model import Model
model = Model('models/ggml-medium.bin', n_threads=6)

>>> model = Model('models/ggml-medium.bin', n_threads=6)
[2024-05-21 11:27:12,237] {model.py:221} INFO - Initializing the model ...
whisper_init_from_file_with_params_no_state: loading model from 'models/ggml-medium.bin'
whisper_init_with_params_no_state: use gpu    = 1
whisper_init_with_params_no_state: flash attn = 0
whisper_init_with_params_no_state: gpu_device = 0
whisper_init_with_params_no_state: dtw        = 0
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51865
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 1024
whisper_model_load: n_audio_head  = 16
whisper_model_load: n_audio_layer = 24
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 1024
whisper_model_load: n_text_head   = 16
whisper_model_load: n_text_layer  = 24
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 4 (medium)
whisper_model_load: adding 1608 extra tokens
whisper_model_load: n_langs       = 99
whisper_backend_init: using Metal backend
ggml_metal_init: allocating
ggml_metal_init: found device: Apple M3 Max
ggml_metal_init: picking default device: Apple M3 Max
ggml_metal_init: default.metallib not found, loading from source
ggml_metal_init: GGML_METAL_PATH_RESOURCES = /Users/cgk/Desktop/code/whisper.cpp/pywhispercpp/whisper.cpp
ggml_metal_init: loading '/Users/cgk/Desktop/code/whisper.cpp/pywhispercpp/whisper.cpp/ggml-metal.metal'
ggml_metal_init: GPU name:   Apple M3 Max
ggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)
ggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)
ggml_metal_init: GPU family: MTLGPUFamilyMetal3  (5001)
ggml_metal_init: simdgroup reduction support   = true
ggml_metal_init: simdgroup matrix mul. support = true
ggml_metal_init: hasUnifiedMemory              = true
ggml_metal_init: recommendedMaxWorkingSetSize  = 51539.61 MB
whisper_model_load:    Metal total size =  1533.14 MB
whisper_model_load: model size    = 1533.14 MB
whisper_init_state: kv self size  =  150.99 MB
whisper_init_state: kv cross size =  150.99 MB
whisper_init_state: kv pad  size  =    6.29 MB
whisper_init_state: loading Core ML model from 'models/ggml-medium-encoder.mlmodelc'
whisper_init_state: first run on a device may take a while ...
whisper_init_state: Core ML model loaded
whisper_init_state: compute buffer (conv)   =    8.81 MB
whisper_init_state: compute buffer (cross)  =    7.85 MB
whisper_init_state: compute buffer (decode) =  142.09 MB
>>> 
>>> print(Model.system_info())
AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | METAL = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | CUDA = 0 | COREML = 1 | OPENVINO = 0
>>> segments = model.transcribe(file, speed_up=True)
[2024-05-21 11:27:58,939] {model.py:130} INFO - Transcribing ...
whisper_full_with_state: failed to compute log mel spectrogram
[2024-05-21 11:27:58,939] {model.py:133} INFO - Inference time: 0.000 s

[absadiki]

@chigkim,
Can you try without speed_up=True:

segments = model.transcribe(file)

[raivisdejus]

I had the same problem on Ubuntu 24.04 and I can confirm that removing speed_up=True fixed the issue

[absadiki]

Okey, thanks @raivisdejus.