whisper : remove whisper_load_backends function #3196
Conversation
|
I did this ages ago but the PR got refused |
|
@danbev The correct fix should be: #3042 (comment) |
This commit removes the `whisper_load_backends` function, which was used to load all GGML backends. The motivation for this change push the responsibility of loading backends to user applications to give them more control over which backends to load and when. See the references below for more context. Resolves: ggml-org#3182 Refs: ggml-org#3042 (comment) Refs: ggml-org#3042 (comment)
danbev
force-pushed
the
backend-dl-load-model
branch
from
4cf2aa3 to
1eb69c0
Compare
danbev
changed the title
Sorry, I've not been following this but trying to catch up now. |
|
You may keep Ruby tests failed if you want to. I will investigate and fix after merged. |
This commit adds a check to ensure that the `rwc` pointer is not NULL before attempting to mark its members in the garbage collector. The motivation for this is an attempt to see if this fixed the CI build as I'm not able to reproduce the issue locally. Refs: https://github.com/ggml-org/whisper.cpp/actions/runs/15299612277/job/43036694928?pr=3196
Thanks, I was not able to reproduce the error locally but made an attempt to "fix" it. But otherwise I will gratefully take you up on your offer :) |
|
There's another problem here... whisper_load_backend calls ggml_backend_load_all This means that cuda would be chosen as the GPU baciend. In my tests #3139 on the same machine Vulkan is faster on Windows and CUDA is faster under Linux - Difference is currently circa 20% May be down to NVIDIA's current driver problems (gap was even worse a few weeks ago) |
This PR is only trying to address the segment fault in the referenced issue. I'm not saying that the other issues are not important just that those should be handled in separate PRs. |
This commit removes the
whisper_load_backendsfunction, which was usedto load all GGML backends.
The motivation for this change push the responsibility of loading
backends to user applications to give them more control over which
backends to load and when. See the references below for more context.
Resolves: #3182
Refs: #3042 (comment)
Refs: #3042 (comment)