mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-04-16 03:26:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

gguf_convert_endian.py: implement byteswapping for q4_k and q6_k (#11349)

Browse Source
This commit is contained in:
Aleksei Nikiforov 2025-02-24 12:27:01 +01:00 committed by GitHub
parent 8303e8b0fb
commit 651adf4b66
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 55 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

55
gguf-py/gguf/scripts/gguf_convert_endian.py
View File

@ -43,6 +43,8 @@ def convert_byteorder(reader: gguf.GGUFReader, args: argparse.Namespace) -> None
gguf.GGMLQuantizationType.F32,
gguf.GGMLQuantizationType.F16,
gguf.GGMLQuantizationType.Q8_0,
gguf.GGMLQuantizationType.Q4_K,
gguf.GGMLQuantizationType.Q6_K,
):
raise ValueError(f"Cannot handle type {tensor.tensor_type.name} for tensor {repr(tensor.name)}")
logger.info(f"* Preparing to convert from {file_endian.upper()} to {order.upper()}")
@ -96,6 +98,59 @@ def convert_byteorder(reader: gguf.GGUFReader, args: argparse.Namespace) -> None
if block_num % 100000 == 0:
inner_pbar.set_description(f"Byte-swapping Blocks [{(n_blocks - block_num) // n_blocks}]")
elif tensor.tensor_type == gguf.GGMLQuantizationType.Q4_K:
# Handle Q4_K tensor blocks (block_q4_k)
# Specific handling of block_q4_k is required.
# Each block_q4_k consists of 2 f16 values followed by 140 int8 values.
# first flatten structure
newshape = 1
for i in tensor.data.shape:
newshape *= i
tensor.data.resize(newshape)
block_size = 144
n_blocks = len(tensor.data) // block_size
for block_num in (inner_pbar := tqdm(range(n_blocks), desc="Byte-swapping Blocks", leave=False)):
block_offs = block_num * block_size
# Byte-Swap f16 sized fields
delta = tensor.data[block_offs:block_offs + 2].view(dtype=np.uint16)
delta.byteswap(inplace=True)
delta = tensor.data[block_offs + 2:block_offs + 4].view(dtype=np.uint16)
delta.byteswap(inplace=True)
# Byte-Swap
if block_num % 100000 == 0:
inner_pbar.set_description(f"Byte-swapping Blocks [{(n_blocks - block_num) // n_blocks}]")
elif tensor.tensor_type == gguf.GGMLQuantizationType.Q6_K:
# Handle Q6_K tensor blocks (block_q6_k)
# Specific handling of block_q6_k is required.
# Each block_q6_k consists of 208 int8 values followed by 1 f16 value.
# first flatten structure
newshape = 1
for i in tensor.data.shape:
newshape *= i
tensor.data.resize(newshape)
block_size = 210
n_blocks = len(tensor.data) // block_size
for block_num in (inner_pbar := tqdm(range(n_blocks), desc="Byte-swapping Blocks", leave=False)):
block_offs = block_num * block_size
# Byte-Swap f16 sized field
delta = tensor.data[block_offs + 208:block_offs + 210].view(dtype=np.uint16)
delta.byteswap(inplace=True)
# Byte-Swap
if block_num % 100000 == 0:
inner_pbar.set_description(f"Byte-swapping Blocks [{(n_blocks - block_num) // n_blocks}]")
else:
# Handle other tensor types
tensor.data.byteswap(inplace=True)