Conversation
- `ggml_rope()` - `ggml_diag_mask_inf()` multi-threaded - compatibility with scratch buffers
Seunghhon
pushed a commit
to Seunghhon/llama.cpp
that referenced
this pull request
Apr 26, 2026
- `ggml_rope()` - `ggml_diag_mask_inf()` multi-threaded - compatibility with scratch buffers
phuongncn
pushed a commit
to phuongncn/llama.cpp-gx10-dgx-sparks-deepseekv4
that referenced
this pull request
Apr 28, 2026
- `ggml_rope()` - `ggml_diag_mask_inf()` multi-threaded - compatibility with scratch buffers
phuongncn
pushed a commit
to phuongncn/llama.cpp-gx10-dgx-sparks-deepseekv4
that referenced
this pull request
Apr 28, 2026
* WIP: mistral4 * CPU FA * CUDA FA 320, 256
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
ggml_rope()when not inplace ggml-org/ggml@788381eggml_rope()GPT-NeoX mode (hopefully) ggml-org/ggml@788381eggml_diag_mask_inf()operator ggml-org/ggml@a483bb2The
ggml_rope()fixes are irrelevant for LLaMA sincen_rot == (n_embd / n_head), but it makes a difference for other models like GPT-J and GPT-NeoX wheren_rot < (n_embd / n_head). I'm still not sure if this is the correct implementation, especially for the GPT-NeoX mode, but results kind of seem a bit better than before.The non-inplace multi-thread
ggml_diag_mask_inf()was broken here: #1428 . Again, irrelevant since in LLaMA forward we useggml_diag_mask_inf_inplace(). Might be relevant to @xaedesThe "scratch buffers" fix might be relevant for LLaMA. See the new
ggml_scratch_save()andggml_scratch_load()functions and their usage inggml.c: https://github.com/ggerganov/llama.cpp/blob/fixes/ggml.c#LL3925C1-L3939C1The scratch buffers are mechanism for reusing memory from previous ops when it is no longer needed. The current way of using them is manual and very error-prone. Will hopefully come up with something better in the future.
More info here: ggml-org/whisper.cpp#431