ROCm AMD Unified Memory Architecture (UMA) handling #4449
ggerganov
merged
7 commits into
ggml-org:master
from
ekg:rocm-amd-uma
AMD ROCm: handle UMA memory VRAM expansions
d59c0b3a
clarify build process for ROCm on linux with cmake
e754a83a
avoid using deprecated ROCm hipMallocHost
405fc540
Merge branch 'master' of https://github.com/ggerganov/llama.cpp into …
6caf33cf
keep simplifying the change required for UMA
7ee8df3d
cmake: enable UMA-compatible allocation when LLAMA_HIP_UMA=ON
1e946c54
Merge branch 'master' of https://github.com/ggerganov/llama.cpp into …
87cfad3c
Reviewers
No reviews
Assignees
No one assigned
Labels
need feedback
Milestone
No milestone
Login to write a write a comment.
Login via GitHub