llama.cpp
93034df7
- iq3_s_mult_shuffle: use lookup table on CUDA
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
1 year ago
iq3_s_mult_shuffle: use lookup table on CUDA ~4% faster TG that way.
References
#5867 - IQ3_S: multiplier based code book
Author
Iwan Kawrakow
Parents
6d15da1e
Loading