llama.cpp
TP: fix 0-sized tensor slices, AllReduce fallback
#21808

Merged

TP: fix 0-sized tensor slices, AllReduce fallback #21808

JohannesGaessler merged 4 commits into ggml-org:master from JohannesGaessler:tp-fix-0-slice

JohannesGaessler requested a review 56 days ago

github-actions added Nvidia GPU

github-actions added ggml

JohannesGaessler requested a review from

CISC 56 days ago

gaugarg-nv commented on 2026-04-13

am17an approved these changes on 2026-04-19

TP: fix 0-sized tensor slices, AllReduce fallback

950907db

fix layer structure <-> GPU count aliasing

7f58ec69

add missing std::fill

9c905258

JohannesGaessler force pushed from 8cd16dc2 to 9c905258 48 days ago

gaugarg-nv commented on 2026-04-20

fix CUDA device set, max ggml ctx size

69f50b19

gaugarg-nv approved these changes on 2026-04-20

JohannesGaessler merged fb19f94c into master 48 days ago

Reviewers

gaugarg-nv

am17an

CISC

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone