biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Kernel/Quant] Remove AQLM (#22943)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Michael Goin
2025-08-16 15:38:21 -04:00
committed by GitHub
parent 3253ae765e
commit 4fc722eca4
16 changed files with 0 additions and 1534 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
tests/kernels/quantization/test_aqlm.py
View File

@@ -1,40 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import torch
from tests.kernels.utils import opcheck
from vllm import _custom_ops as ops # noqa: F401
def test_aqlm_dequant_opcheck():
codes = torch.randint(-32768,
32767, (22016, 512, 1),
device='cuda',
dtype=torch.int16)
codebooks = torch.rand((2, 65536, 1, 8),
device='cuda',
dtype=torch.float16)
codebook_partition_sizes = [11008, 11008]
opcheck(torch.ops._C.aqlm_dequant,
(codes, codebooks, codebook_partition_sizes))
def test_aqlm_gemm_opcheck():
input = torch.rand((4, 4096), device='cuda', dtype=torch.float16)
codes = torch.randint(-32768,
32767, (12288, 512, 1),
device='cuda',
dtype=torch.int16)
codebooks = torch.rand((3, 65536, 1, 8),
device='cuda',
dtype=torch.float16)
scales = torch.rand((12288, 1, 1, 1), device='cuda', dtype=torch.float16)
codebook_partition_sizes = [4096, 4096, 4096]
bias = None
opcheck(torch.ops._C.aqlm_gemm,
(input, codes, codebooks, scales, codebook_partition_sizes, None))
opcheck(torch.ops._C.aqlm_gemm,
(input, codes, codebooks, scales, codebook_partition_sizes, bias))