PR #1913 Convert vector to f16 for dequantize mul mat vec

Convert vector to f16 for dequantize mul mat vec #1913

JohannesGaessler merged 10 commits into ggml-org:master from JohannesGaessler:cuda-dmmv-f16-3

JohannesGaessler requested a review from

ikawrakow 2 years ago

JohannesGaessler added performance

JohannesGaessler added hardware

Convert vector to f16 for dmmv

98b8596b

dfloat2

8ac993bd

compile option

cde81f91

works

28e27623

ikawrakow requested changes on 2023-06-18

WIP better data types

e6c9b5f5

cleanup

a2db9896

JohannesGaessler force pushed from f66c095b to a2db9896 2 years ago

cmake compilation options

76caff6e

Added compilation option description to README

d036161b

Fix trailing whitespace

a794320f

Fix cmake compilation issues

621d6264

ikawrakow requested a review from

ikawrakow 2 years ago

ikawrakow approved these changes on 2023-06-18

JohannesGaessler merged 16b9cd19 into master 2 years ago

Reviewers

ikawrakow

Assignees

No one assigned

Labels

performance hardware

Milestone

No milestone