llama.cpp
llama-graph: prevent f16 overflow in MoE down-projection on Metal
#25389
Closed
Closed
llama-graph: prevent f16 overflow in MoE down-projection on Metal #25389
joerowell
wants to merge
1 commit into
ggml-org:master
from
joerowell:metal-moe-down-f16-overflow
llama-graph: prevent f16 overflow in MoE down-projection on Metal
ffd31cf3
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub