xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 03:44:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CI Failure] Fix torchao dep failure for Quantization Test (#26824)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin 2025-10-14 19:41:43 -04:00 committed by GitHub
parent 4aed506b65
commit 7e0ef4084a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 8 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.buildkite/test-amd.yaml
View File

@ -603,7 +603,8 @@ steps:
# since torchao nightly is only compatible with torch nightly currently
# https://github.com/pytorch/ao/issues/2919, we'll have to skip new torchao tests for now
# we can only upgrade after this is resolved
- pip install --pre torchao==0.13.0.dev20250814 --index-url https://download.pytorch.org/whl/nightly/cu128
# TODO(jerryzh168): resolve the above comment
- uv pip install --system torchao==0.13.0
- VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/
- label: LM Eval Small Models # 53min

3
.buildkite/test-pipeline.yaml
View File

@ -527,7 +527,8 @@ steps:
# since torchao nightly is only compatible with torch nightly currently
# https://github.com/pytorch/ao/issues/2919, we'll have to skip new torchao tests for now
# we can only upgrade after this is resolved
- pip install --pre torchao==0.13.0.dev20250814 --index-url https://download.pytorch.org/whl/nightly/cu128
# TODO(jerryzh168): resolve the above comment
- uv pip install --system torchao==0.13.0
- VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/
- label: LM Eval Small Models # 53min

3
tests/quantization/test_compressed_tensors.py
View File

@ -697,7 +697,8 @@ def test_compressed_tensors_2of4_sparse_compressed(vllm_runner, args_2of4):
@pytest.mark.parametrize(
"args",
[
("nm-testing/TinyLlama-1.1B-Chat-v1.0-NVFP4A16", CompressedTensorsW4A16Fp4),
# TODO: Enable once model is available again
# ("nm-testing/TinyLlama-1.1B-Chat-v1.0-NVFP4A16", CompressedTensorsW4A16Fp4),
("nm-testing/TinyLlama-1.1B-Chat-v1.0-NVFP4", CompressedTensorsW4A4Fp4),
],
)

3
vllm/model_executor/layers/quantization/rtn.py
View File

@ -15,6 +15,7 @@ from vllm.model_executor.layers.fused_moe.config import (
FusedMoEConfig,
FusedMoEQuantConfig,
)
from vllm.model_executor.layers.fused_moe.fused_marlin_moe import fused_marlin_moe
from vllm.model_executor.layers.fused_moe.layer import FusedMoE, FusedMoEMethodBase
from vllm.model_executor.layers.linear import (
LinearBase,
@ -396,7 +397,7 @@ class RTNMoEMethod(FusedMoEMethodBase):
indices_type=self.topk_indices_dtype,
)
return torch.ops.vllm.fused_marlin_moe(
return fused_marlin_moe(
x,
layer.w13_weight,
layer.w2_weight,