]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 32b741c)
sycl: Increase minimum buffer size for USM system allocations (#25525)
authorFrancois Dugast <redacted>
Wed, 15 Jul 2026 07:28:24 +0000 (09:28 +0200)
committerGitHub <redacted>
Wed, 15 Jul 2026 07:28:24 +0000 (10:28 +0300)
commit0e148a573f0ccbb0e9000312e2757b8998c1e070
tree2ebc23e1543feb504a3353e017d5551cb764ab2ftree
parent32b741c336decea914e4c1c24a9c9815485901b2commit | diff
sycl: Increase minimum buffer size for USM system allocations (#25525)

Raise the threshold for minimum buffer size from 1 GiB to 4 GiB, based
on real-world experiments of overcommitting device memory with model
weights larger than available VRAM, for example Qwen3.5-35B-A3B-Q8
running on a B70.

Also add a debug message to better track USM system allocations.

Signed-off-by: Francois Dugast <redacted>
ggml/src/ggml-sycl/ggml-sycl.cpp diff | blob | history
Packaging of ggml-org/llama.cpp