Skip to content

CPU: workaround avx512 4bit dequantize accuracy issue for large blocksize #1828

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
matthewdouglas merged 1 commit into from
Dec 10, 2025
Merged

CPU: workaround avx512 4bit dequantize accuracy issue for large blocksize #1828

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 9 additions & 1 deletion bitsandbytes/backends/cpu/ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,8 @@

logger = logging.getLogger(__name__)

_has_avx512 = torch.backends.cpu.get_cpu_capability() == "AVX512"

# torch._int_mm for s8@s8->s32 is supported on CPU from torch 2.4+.
# However, we can overflow if we use this without AVX512_VNNI support.
# This is fixed in torch 2.6+, so we set this as the minimum to be safe.
Expand Down Expand Up @@ -134,8 +136,14 @@ def _(
lambda: f"Blockwise 4bit dequantization only supports 16/32-bit floats, but got {dtype}",
)

# Fallback as AVX512 implementation has accuracy issues with fp16/fp32 and blocksize >= 2048
# Note: this is not a common use case.
avx512_fallback = _has_avx512 and blocksize >= 2048 and dtype != torch.bfloat16

# Odd shape is not supported by this kernel; fallback to generic implementation
if shape[-1] % 2 != 0:
shape_fallback = shape[-1] % 2 != 0

if avx512_fallback or shape_fallback:
from ..default.ops import _dequantize_4bit_impl

return _dequantize_4bit_impl(A, absmax, blocksize, quant_type, shape, dtype)
Expand Down