PR #15434 llama: use FA + max. GPU layers by default - SemanticDiff

Commits

llama: use max. GPU layers by default, auto -fa

JohannesGaessler committed 225 days ago
disable -fa for server test

JohannesGaessler committed 225 days ago
remove redundant defaults

JohannesGaessler committed 223 days ago
ggml-backend: abort instead of segfault

JohannesGaessler committed 223 days ago
address review comments

JohannesGaessler committed 222 days ago
add comment [no ci]

JohannesGaessler committed 222 days ago
fix unittest, remove metal ifdef

JohannesGaessler committed 222 days ago
add comment [no ci]

JohannesGaessler committed 222 days ago

Loading