ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 03:11:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

bitnet: add 2 bpw quantization

Browse Source 
The scalar dot product already chieves 37 t/s for TG!
This commit is contained in:
Kawrakow
2024-06-17 18:41:30 +03:00
parent f9ba085ef7
commit 318899c8b7
10 changed files with 216 additions and 184 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml.h
View File

@@ -384,7 +384,8 @@ extern "C" {
GGML_TYPE_IQ1_M = 29,
GGML_TYPE_BF16 = 30,
GGML_TYPE_IQ1_BN = 31,
GGML_TYPE_Q8_K64 = 32,
GGML_TYPE_IQ2_BN = 32,
GGML_TYPE_Q8_K64 = 33,
GGML_TYPE_COUNT,
};
@@ -427,6 +428,7 @@ extern "C" {
GGML_FTYPE_MOSTLY_IQ1_M = 23, // except 1d tensors
GGML_FTYPE_MOSTLY_BF16 = 24, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ1_BN = 25, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ2_BN = 26, // except 1d tensors
};
// available tensor operations: