You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi! First thing first thanks for your work, this is an awesome tool if one has some metal to spare and doesn't want to give 23 bucks per year to TG.
I have encountered the error mentioned in issue #7 too and I think it may be related to this: ggerganov/whisper.cpp#83
It seems that the CPU lacks specific instructions invoked by Whisper (which I guess may happen if the CPU is old enough - in my case its almost 12 years old). Probably it's a non-AVX CPU.
A possible solution (albeit to test yet) to enable retrocompatibility could be to allow local compilation of the model (maybe via .env variables) and to run Cmake with the -DWHISPER_NO_SIMD=ON parameter.
EDIT 1:
Actually per tazz4843/whisper-rs#8 it seems ggerganov already updated their Makefile to handle this at compilation time, so I suppose it would boil down to allowing local compilation.
EDIT 2: To give some specs, I tried with multiple models, from tiny to large v3 quantized, but got the same error.
As I mentioned, the machine I used for this is quite old (it is this laptop), currently running Xubuntu 24 LTS x64.
The process successfully converts the audio file via FFMPEG, but dies with the Illegal Instruction (Core dumped) exception when running Whisper.
The text was updated successfully, but these errors were encountered:
frnprt
changed the title
[Errno 2] No such file or directory
[Errno 2] No such file or directory - Illegal Instruction (core dumped)
Dec 4, 2024
Hi! First thing first thanks for your work, this is an awesome tool if one has some metal to spare and doesn't want to give 23 bucks per year to TG.
I have encountered the error mentioned in issue #7 too and I think it may be related to this:
ggerganov/whisper.cpp#83
It seems that the CPU lacks specific instructions invoked by Whisper (which I guess may happen if the CPU is old enough - in my case its almost 12 years old). Probably it's a non-AVX CPU.
A possible solution (albeit to test yet) to enable retrocompatibility could be to allow local compilation of the model (maybe via .env variables) and to run Cmake with the
-DWHISPER_NO_SIMD=ON
parameter.EDIT 1:
Actually per tazz4843/whisper-rs#8 it seems ggerganov already updated their Makefile to handle this at compilation time, so I suppose it would boil down to allowing local compilation.
EDIT 2: To give some specs, I tried with multiple models, from tiny to large v3 quantized, but got the same error.
As I mentioned, the machine I used for this is quite old (it is this laptop), currently running Xubuntu 24 LTS x64.
The process successfully converts the audio file via FFMPEG, but dies with the
Illegal Instruction (Core dumped)
exception when running Whisper.The text was updated successfully, but these errors were encountered: