llama.cpp
metal : make the backend async v2
#15906
Merged

metal : make the backend async v2 #15906

ggerganov merged 17 commits into master from gg/metal-async-v2
ggerganov
ggerganov metal : make the backend async
97b96c1a
ggerganov cont : add comments, extend op offload, clean up
c5637cf3
ggerganov metal : fix batch size for MUL_MAT_ID
bdff7729
ggerganov metal : remove deprecated ggml_backend_metal_buffer_from_ptr
d91ba85d
ggerganov metal : create only metal buffers, no wrapping of host memory
85aaf52b
ggerganov metal : restore .alloc_buffer for buffer_from_ptr_type
7fc2b3d5
ggerganov metal : remove broken implementation of GGML_OP_SET
f288225d
ggerganov metal : clean-up loose ends, ready for tests
0926cb49
ggerganov metal : support both private and shared buffers
7b59f0f7
slaren
slaren
ggerganov
github-actions github-actions added ggml
github-actions github-actions added Apple Metal
slaren
slaren
ggerganov
ggerganov metal : enable private buffers + add global device queue
040844cf
ggerganov metal : disable host buffer to prevent races
afd95d20
ggerganov metal : avoid extra copy during set_tensor
e65c53e9
ggerganov
slaren
ggerganov metal : use separate buffer types for shread and private Metal buffers
523750a8
ggerganov metal : simplify synchronization logic
9248aecd
ggerganov
ggerganov commented on 2025-09-10
ggerganov
ggerganov metal : fix build
5fdce91b
slaren
slaren commented on 2025-09-10
slaren
ggerganov metal : do not implement cpy_tensor
c9a5ba47
ggerganov metal : separate implementations for shared and private buffers
e796f669
ggerganov
ggerganov ggerganov merged 0f0a3c28 into master 30 days ago
ggerganov ggerganov deleted the gg/metal-async-v2 branch 30 days ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone