Repository navigation
ggml : add BF16 unary, GLU, binary and scale ops (CPU, CUDA) - #29675
Merged
Merged
Conversation
ggerganov
reviewed
Sep 29, 2026
Comment on lines
+2701
to
+2704
| case GGML_TYPE_BF16: | ||
| { | ||
| ggml_compute_forward_unary_bf16(params, dst); | ||
| } break; |
Member
There was a problem hiding this comment.
This should be ggml_compute_forward_silu_bf16 to follow the existing pattern. Otherwise, we should consolidate the f16 and f32 paths in a similar "unary" path.
github-actions
Bot
added
testing
JohannesGaessler
approved these changes
Sep 29, 2026
JohannesGaessler
left a comment
JohannesGaessler
left a comment
Contributor
There was a problem hiding this comment.
The CUDA changes LGTM. In principle it should be possible to improve performance further by loading more than 2 bytes per thread at once.
ggerganov
approved these changes
Sep 30, 2026
pierreguillot
pushed a commit
to Ircam-Partiels/llama.cpp
that referenced
this pull request
Oct 1, 2026
…g#29675) * ggml : add BF16 unary, GLU, binary and scale ops (CPU, CUDA) * ggml-cpu : use per-op _bf16 functions for BF16 unary and GLU ops Assisted-by: Claude Opus 5.5 * CUDA: use ggml_cuda_cast in binbcast and unary kernels to fix the HIP bf16 build * ggml-openvino : reject BF16 SCALE and mixed-type BF16 ADD/MUL/SUB
cwriter
pushed a commit
to cwriter/llama.cpp
that referenced
this pull request
Oct 4, 2026
The kernels assert F32/F16 (GLU) and F32 (SCALE), so the BF16 cases that test-backend-ops gained with ggml-org#29675 aborted the run. Assisted-by: Claude Opus 5.5
|
@am17an thanks for this. We're interested in the BF16-activations follow-up you mentioned. We run a small vision-transformer world model on ggml with large batches and tiny weights, so we're bound by activation traffic. A BF16 result from |
1 task
frostyautumnleaf
pushed a commit
to frostyautumnleaf/llama.cpp
that referenced
this pull request
Oct 5, 2026
…g#29675) * ggml : add BF16 unary, GLU, binary and scale ops (CPU, CUDA) * ggml-cpu : use per-op _bf16 functions for BF16 unary and GLU ops Assisted-by: Claude Opus 5.5 * CUDA: use ggml_cuda_cast in binbcast and unary kernels to fix the HIP bf16 build * ggml-openvino : reject BF16 SCALE and mixed-type BF16 ADD/MUL/SUB
Wizard815
pushed a commit
to Wizard815/mx-llama.cpp-Rocm10
that referenced
this pull request
Oct 6, 2026
…g#29675) * ggml : add BF16 unary, GLU, binary and scale ops (CPU, CUDA) * ggml-cpu : use per-op _bf16 functions for BF16 unary and GLU ops Assisted-by: Claude Opus 5.5 * CUDA: use ggml_cuda_cast in binbcast and unary kernels to fix the HIP bf16 build * ggml-openvino : reject BF16 SCALE and mixed-type BF16 ADD/MUL/SUB (cherry picked from commit 2090f60)
edwardyoon
pushed a commit
to edwardyoon/focus-llama
that referenced
this pull request
Oct 7, 2026
…g#29675) * ggml : add BF16 unary, GLU, binary and scale ops (CPU, CUDA) * ggml-cpu : use per-op _bf16 functions for BF16 unary and GLU ops Assisted-by: Claude Opus 5.5 * CUDA: use ggml_cuda_cast in binbcast and unary kernels to fix the HIP bf16 build * ggml-openvino : reject BF16 SCALE and mixed-type BF16 ADD/MUL/SUB (cherry picked from commit 2090f60)
|
Is this the PR that increases quality for Gemma 4 QAT or will it be the upcoming BF16 activations PR that is based on this? |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
Overview
Add bf16 variants of element-wise ops. The reason for adding these is to keep activations in bf16 (which will added for a later PR). Anyway I think it doesn't hurt to have these.
Additional information
Requirements