ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-10 05:50:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Adding IQ5_KS - 5.25 bpw quants (#422)

Browse Source 
* iq5_ks: basics

* iq5_ks: quantize

* iq5_ks: CUDA dequantize works

* iq5_ks: dot product works on CUDA

* iq5_ks: MMQ works

* iq5_ks: Zen4

* iq5_ks: AVX2

But is is not quite right, just like iq4_k, iq5_k, iq6_k, iq4_ks.
All these need fixing on AVX2.

* iq5_ks: NEON

* iq5_ks: Metal dequantize

* iq5_ks: Metal dot product

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-05-15 16:02:39 +03:00
committed by GitHub
parent 17d721820a
commit 90e53a0b8b
20 changed files with 848 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
include/llama.h
View File

@@ -193,6 +193,7 @@ extern "C" {
LLAMA_FTYPE_MOSTLY_IQ2_KS = 147, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ4_KSS = 148, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q8_KV = 149, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ5_KS = 150, // except 1d tensors
//
LLAMA_FTYPE_MOSTLY_Q4_0_R8 = 202, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q8_0_R8 = 207, // except 1d tensors
@@ -231,7 +232,7 @@ extern "C" {
LLAMA_ROPE_SCALING_TYPE_LINEAR = 1,
LLAMA_ROPE_SCALING_TYPE_YARN = 2,
LLAMA_ROPE_SCALING_TYPE_LONGROPE = 3,
LLAMA_ROPE_SCALING_TYPE_MAX_VALUE = LLAMA_ROPE_SCALING_TYPE_LONGROPE,
LLAMA_ROPE_SCALING_TYPE_MAX_VALUE = LLAMA_ROPE_SCALING_TYPE_LONGROPE,
};
enum llama_pooling_type {