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
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The reason will be displayed to describe this comment to others. Learn more.
This commit specifically breaks CLBlast for me. It just segfaults upon trying to initialize CLBlast and generating text. python koboldcpp.py '../ggml-pythia-160m-deduped-q4_3.bin' --threads 6 --stream --smartcontext --useclblast 0 0
Processing Prompt [BLAS] (256 / 580 tokens)
Initializing CLBlast (First Run)...
Attempting to use: Platform=0, Device=0 (If invalid, program will crash)
Segmentation fault (core dumped)
Does not effect e97c709. python koboldcpp.py '../ggml-pythia-160m-deduped-q4_3.bin' --threads 6 --stream --smartcontext --useclblast 0 0
The reason will be displayed to describe this comment to others. Learn more.
No, I tried getting rid of that line, and if I git clone and checkout the commit before this, it seg faults. The only way I don't get a segfault is to download the files for the commit from the permalink at that point of history in the repo.
The reason will be displayed to describe this comment to others. Learn more.
This makes no sense, the files are completely identical.
Edit: dragging and dropping all the files from the non-working folder to the working folder and replacing them = no seg fault.. then dragging them back to the folder that doesn't work.. seg fault.
The reason will be displayed to describe this comment to others. Learn more.
Sounds like something is wonky with your environment or filesystem, if the files are identical. Could it be some directory permissions you set? Might want to do a clean git clone to a different directory.
I'd also try copying the directory and pruning the git files. Maybe they are messing with it somehow.
The reason will be displayed to describe this comment to others. Learn more.
I think it's something wonky with how things link with the opencl-amd package on Arch..
I switched to opencl-mesa and now it works, not only that, but it provides better performance than the AMD package for OpenCL. (50ms down to 40ms on a 7B model)
🤷
Think I'm going to do a clean install either way.
The reason will be displayed to describe this comment to others. Learn more.
Reinstalled Arch, gonna figure it was a fluke, though I'm using opencl-mesa for CLBlast now as it provides better performance, and apparently it will gain more speed soon-ish.
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This commit specifically breaks CLBlast for me. It just segfaults upon trying to initialize CLBlast and generating text.
python koboldcpp.py '../ggml-pythia-160m-deduped-q4_3.bin' --threads 6 --stream --smartcontext --useclblast 0 0
Does not effect e97c709.
python koboldcpp.py '../ggml-pythia-160m-deduped-q4_3.bin' --threads 6 --stream --smartcontext --useclblast 0 0
Honestly it doesn't make sense. Commenting out the added line doesn't fix it.
Edit: Even stranger, if I browse the files of this commit and download those files instead of the latest files, it works.
https://github.com/LostRuins/koboldcpp/archive/refs/heads/concedo.zip Seg faults.
https://github.com/LostRuins/koboldcpp/archive/f75de52b25b9e4ae8acb7c751521d624ff360e87.zip Works fine.
git clone https://github.com/LostRuins/koboldcpp
Seg faults.f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That is very strange. Nothing in this update should have affected CLBlast at all
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you compare the files in your checkout repo vs the zip? any difference?
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It makes zero sense. All that this commit has changed is adding one line of sleep function. Commenting it does nothing?
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No, I tried getting rid of that line, and if I git clone and checkout the commit before this, it seg faults. The only way I don't get a segfault is to download the files for the commit from the permalink at that point of history in the repo.
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
According to meld, the folder that works and the folder that doesn't are 100% the same with no differences..
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This makes no sense, the files are completely identical.
Edit: dragging and dropping all the files from the non-working folder to the working folder and replacing them = no seg fault.. then dragging them back to the folder that doesn't work.. seg fault.
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sounds like something is wonky with your environment or filesystem, if the files are identical. Could it be some directory permissions you set? Might want to do a clean git clone to a different directory.
I'd also try copying the directory and pruning the git files. Maybe they are messing with it somehow.
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think it's something wonky with how things link with the opencl-amd package on Arch..
I switched to
opencl-mesa
and now it works, not only that, but it provides better performance than the AMD package for OpenCL. (50ms down to 40ms on a 7B model)🤷
Think I'm going to do a clean install either way.
f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reinstalled Arch, gonna figure it was a fluke, though I'm using
opencl-mesa
for CLBlast now as it provides better performance, and apparently it will gain more speed soon-ish.f75de52
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great, glad you got it fixed