vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-06-01 06:00:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

convert : minor fixes for numpy 2.x (#23571)

Browse source
This commit is contained in:
Sigbjørn Skjæret 2026-05-24 09:51:31 +02:00 • committed by GitHub
parent 63248fc3e3
commit 5d246a792d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

3
examples/convert_legacy_llama.py
View file

@ -1308,7 +1308,8 @@ def do_dump_model(model_plus: ModelPlus) -> None:
def main(args_in: list[str] | None = None) -> None:
output_choices = ["f32", "f16"]
if np.uint32(1) == np.uint32(1).newbyteorder("<"):
dummy_val = np.uint32(1)
if dummy_val == dummy_val.view(dummy_val.dtype.newbyteorder("<")):
# We currently only support Q8_0 output on little endian systems.
output_choices.append("q8_0")
parser = argparse.ArgumentParser(description="Convert a LLaMA model to a GGML compatible file")

1
gguf-py/gguf/quants.py
View file

@ -28,6 +28,7 @@ def quant_shape_from_byte_shape(shape: Sequence[int], quant_type: GGMLQuantizati
# This is faster than np.vectorize and np.apply_along_axis because it works on more than one row at a time
def _apply_over_grouped_rows(func: Callable[[np.ndarray], np.ndarray], arr: np.ndarray, otype: DTypeLike, oshape: tuple[int, ...]) -> np.ndarray:
rows = arr.reshape((-1, arr.shape[-1]))
assert len(rows.shape)
osize = 1
for dim in oshape:
osize *= dim