vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : add Q3_K_XS (#5060)

Browse source 
* Add Q3_K_XS - intermediate size between Q2_K and Q3_K_S

* Q3_K_XS: quanize first 1/8 of ffn_down layers with Q4_K

Together with an importance matrix, this brings perplexity
for LLaMA-v2-70B below the perplexity of the former Q2_K
with a 800 MB smaller quantized model size.

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow 2024-01-22 12:43:33 +02:00 • committed by GitHub
parent 57744932c6
commit 66d575c45c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 48 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

1
llama.h
View file

@ -107,6 +107,7 @@ extern "C" {
LLAMA_FTYPE_MOSTLY_IQ2_XXS = 19, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ2_XS = 20, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q2_K_S = 21, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_XS = 22, // except 1d tensors
LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
};