llama : support GEGLU for jina-bert-v2 #14090
Merged
Conversation
This file contains hidden or 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
ggml-ci
ggerganov
reviewed
Jun 10, 2025
src/llama-graph.cpp
Outdated
Comment on lines
666
to
667
| ggml_tensor * x0 = ggml_view_2d(ctx0, cur, split_point, cur->ne[1], cur->nb[1], 0); | ||
| ggml_tensor * x1 = ggml_view_2d(ctx0, cur, split_point, cur->ne[1], cur->nb[1], split_point * ggml_element_size(cur)); |
There was a problem hiding this comment.
Removing the conts works because the ops will be offloaded back to the CPU if the tensor is not contiguous:
llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu
Lines 2972 to 2993 in 2bb0467
For now lets keep the conts and expand the TODO to reference this comment, so that we update the backend implementations to support those.
There was a problem hiding this comment.
@ggerganov Just to make sure I'm on the right track here, I started looking into this yesterday, rather than (or maybe in addition to) making these ops handle non-contiguous data it seems much simpler (and more efficient) to add GEGLU/SWIGLU ops.
ggerganov
approved these changes
Jun 10, 2025
gabe-l-hart
added a commit
to gabe-l-hart/llama.cpp
that referenced
this pull request
Jun 10, 2025
* origin/master: llama : support GEGLU for jina-bert-v2 (ggml-org#14090) vulkan: force device 0 in CI (ggml-org#14106) Fixed spec timings to: accepted/tested instead of accepted/drafted (ggml-org#14104) sync : ggml ggml : fix weak alias win32 (whisper/0) Vulkan: Don't default to CPU device (like llvmpipe), even if no other device is available, to allow fallback to CPU backend (ggml-org#14099) rpc : nicer error messages for RPC server crash (ggml-org#14076) sync : ggml Add in-build ggml::ggml ALIAS library (ggml/1260) metal : use less stack memory in FA kernel (ggml-org#14088) kv-cache : fix shift and defrag logic (ggml-org#14081) llama : allow building all tests on windows when not using shared libs (ggml-org#13980)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Clean up conversion of
jina-bert-v2models now that we haveLLM_FFN_GEGLU, as we can easily support both old and newly converted models.Removedggml_continLLM_FFN_SWIGLUandLLM_FFN_GEGLU, they are unnecessary.Edit: Tested withoutggml_conton CPU and CUDA withjina-reranker-v1(GEGLU) andPhi-3.5-mini-instruct(SWIGLU).