Fixed mmap prefetch for GPU offloading#2529
Merged
JohannesGaessler merged 1 commit intoggml-org:masterfrom Aug 7, 2023
Merged
Conversation
ggerganov
approved these changes
Aug 6, 2023
slaren
reviewed
Aug 6, 2023
JohannesGaessler
force-pushed
the
cuda-fix-mmap-prefetch
branch
from
9648436 to
d9024df
Compare
slaren
approved these changes
Aug 6, 2023
Seunghhon
pushed a commit
to Seunghhon/llama.cpp
that referenced
this pull request
Apr 26, 2026
phuongncn
pushed a commit
to phuongncn/llama.cpp-gx10-dgx-sparks-deepseekv4
that referenced
this pull request
Apr 28, 2026
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
4 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
I've noticed that currently when loading models to VRAM (with mmap) the entire model first gets loaded into RAM and then the model get loaded from RAM to VRAM. This is not how I intended it since this will very likely cause problems when running models with a total size larger than RAM. Instead the loading should be lazy so the model file only needs to be iterated over once (ideally with the CPU tensors being mlocked). Looking through the code I've noticed that when the prefetch size is > 0 the
MAP_POPULATEflag gets set unconditionally; this flag eagerly loads the entire file into memory and it seems I forgot to adjust the logic when I changedprefetchfrombooltosize_t. This PR fixes the logic to only eagerly load the file if the requested prefetch is at least as big as the file size.