upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-06 10:00:48 -05:00
Code Issues Packages Projects Releases Wiki Activity

convert: Add endianness conversion for Q1 and TQ2 quantizations (#26618)

Browse Source 
* Add endianness conversion for Q1 and TQ2 quantizations

* lint

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
This commit is contained in:
Andreas Krebbel
2026-08-05 12:06:09 +02:00
committed by GitHub
parent 61881b1f7f
commit 020760adfc
1 changed files with 18 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
gguf-py/gguf/scripts/gguf_convert_endian.py
View File

@@ -59,11 +59,29 @@ def byteswap_q6_k(tensor, block_offs):
delta.byteswap(inplace=True)
def byteswap_q1_0(tensor, block_offs):
# Each block_q1_0 consists of an f16 delta followed by 16 int8 quantizations.
# Byte-Swap f16 sized delta field
delta = tensor.data[block_offs:block_offs + 2].view(dtype=np.uint16)
delta.byteswap(inplace=True)
def byteswap_tq2_0(tensor, block_offs):
# Each block_tq2_0 consists of 64 int8 values followed by 1 f16 value.
# Byte-Swap f16 sized field
delta = tensor.data[block_offs + 64:block_offs + 66].view(dtype=np.uint16)
delta.byteswap(inplace=True)
byteswap_tensors = {
gguf.GGMLQuantizationType.Q1_0: byteswap_q1_0,
gguf.GGMLQuantizationType.Q4_0: byteswap_q4_0,
gguf.GGMLQuantizationType.Q8_0: byteswap_q8_0,
gguf.GGMLQuantizationType.Q4_K: byteswap_q4_k,
gguf.GGMLQuantizationType.Q6_K: byteswap_q6_k,
gguf.GGMLQuantizationType.TQ2_0: byteswap_tq2_0,
gguf.GGMLQuantizationType.MXFP4: byteswap_noop,
gguf.GGMLQuantizationType.NVFP4: byteswap_noop,
}