Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

runs perfectly with the regular models, but not the quantized ones #993

Open
patrickfconnolly opened this issue Jun 7, 2023 · 8 comments

Comments

@patrickfconnolly
Copy link

patrickfconnolly commented Jun 7, 2023

I have run quantization on ggml-small.en.bin to produce ggml-small.en.bin-q5_0.bin
Quantization proceeded without any errors.

When I run the model, it attempts to load, but throws the following:
GGML_ASSERT: ggml.c:4288: wtype != GGML_TYPE_COUNT
Abort trap: 6

Same issue occurs when I try 8-bit quantization.

Running on a 2015 Macbook Air, if this is any use.

@fieldplower
Copy link

Ye same thing over here, sad hours

@StaffanBetner
Copy link

I get the same error. I use Windows 10 with this version: whisper-bin-x64.zip

>main.exe -f input.wav -l auto -m models\ggml-model-whisper-small.en-q5_1.bin
whisper_init_from_file_no_state: loading model from 'models\ggml-model-whisper-small.en-q5_1.bin'
whisper_model_load: loading model
GGML_ASSERT: D:\a\whisper.cpp\whisper.cpp\ggml.c:4213: wtype != GGML_TYPE_COUNT

@ggerganov
Copy link
Owner

What happens if you build from source using latest master
Make sure to make clean first

@teddybear082
Copy link

Will there be a new official release to support the breaking change to the ggml models that occurred since April 30 by any chance? I get the same error as StaffanBetner with the 1.4.0 windows 64 exe release and downloading the q5 model from ggml.ggerganov so I assume this is the reason. Thanks for all your work on this project! Its amazing.

@StaffanBetner
Copy link

What happens if you build from source using latest master Make sure to make clean first

I downloaded 1.4.2 from here: https://github.com/ggerganov/whisper.cpp/actions/runs/4973278607 and it works with the quantized models. Maybe it is a good idea to highlight in the readme that there are autocompiled versions under the action tab.

@teddybear082
Copy link

Maybe it is a good idea to highlight in the readme that there are autocompiled versions under the action tab

Wow thanks, yes, I had no idea! appreciate you letting us know.

@tazz4843
Copy link
Contributor

tazz4843 commented Jul 2, 2023

Can reproduce this using CuBLAS on a RTX 3060 using v1.4.2. I don't want to use master if possible, so a new release to wrap up any changes since would be appreciated.

[zero@archlinux stt-service-whisper]$ ./target/release/scripty_stt_service /opt/whisper_models/ggml-base-q4_0.bin 4
2023-07-01T23:52:15.291328Z  INFO scripty_stt_service: loading models
2023-07-01T23:52:15.291340Z  INFO stts_speech_to_text: attempting to load model
whisper_init_from_file_no_state: loading model from '/opt/whisper_models/ggml-base-q4_0.bin'
whisper_model_load: loading model
GGML_ASSERT: /home/zero/.cargo/git/checkouts/whisper-rs-c8b9fa7c09c8c913/26c5974/sys/whisper.cpp/ggml.c:4213: wtype != GGML_TYPE_COUNT
Aborted (core dumped)

@ggerganov
Copy link
Owner

Will probably make a new release after merging #1058

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

6 participants