ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 09:09:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add MXFP4 to gguf-py constants (#1007)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-11-24 15:43:49 +01:00
committed by GitHub
parent a3b8efd687
commit 9337229274
1 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
gguf-py/gguf/constants.py
View File

@@ -1468,6 +1468,7 @@ class GGMLQuantizationType(IntEnum):
Q4_0_4_8 = 32
Q4_0_8_8 = 33
I2_S = 36
MXFP4 = 39
Q8_0_X4 = 97
Q8_1_X4 = 98
Q8_2_X4 = 99
@@ -1559,9 +1560,10 @@ class LlamaFileType(IntEnum):
MOSTLY_IQ4_XS = 22 #except 1d tensors
MOSTLY_IQ1_M = 23 #except 1d tensors
MOSTLY_BF16 = 24 #except 1d tensors
MOSTLY_Q4_0_4_4 = 25 #except 1d tensors
MOSTLY_Q4_0_4_8 = 26 #except 1d tensors
MOSTLY_Q4_0_8_8 = 27 #except 1d tensors
MOSTLY_MXFP4 = 25 #except 1d tensors
MOSTLY_Q4_0_4_4 = 26 #except 1d tensors
MOSTLY_Q4_0_4_8 = 27 #except 1d tensors
MOSTLY_Q4_0_8_8 = 28 #except 1d tensors
MOSTLY_Q6_0 = 127 #except 1d tensors
MOSTLY_IQ1_BN = 128 #except 1d tensors
MOSTLY_IQ2_BN = 129 #except 1d tensors
@@ -1682,6 +1684,7 @@ GGML_QUANT_SIZES: dict[GGMLQuantizationType, tuple[int, int]] = {
GGMLQuantizationType.F64 : ( 1, 8),
GGMLQuantizationType.IQ1_M : ( 256, 56),
GGMLQuantizationType.BF16 : ( 1, 2),
GGMLQuantizationType.MXFP4 : ( 32, 17),
GGMLQuantizationType.Q4_0_4_4 : ( 32, 18),
GGMLQuantizationType.Q4_0_4_8 : ( 32, 18),
GGMLQuantizationType.Q4_0_8_8 : ( 32, 18),