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

[CI/Build][REDO] Add is_quant_method_supported to control quantization test configurations (#5466)

Browse Source
This commit is contained in:
Michael Goin
2024-06-13 11:18:08 -04:00
committed by GitHub
parent c2637a613b
commit 23ec72fa03
8 changed files with 32 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tests/quantization/test_bitsandbytes.py
View File

@@ -5,16 +5,12 @@ Run `pytest tests/quantization/test_bitsandbytes.py`.
import pytest
import torch
from tests.quantization.utils import is_quant_method_supported
from vllm import SamplingParams
from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
capability = torch.cuda.get_device_capability()
capability = capability[0] * 10 + capability[1]
@pytest.mark.skipif(
capability < QUANTIZATION_METHODS['bitsandbytes'].get_min_capability(),
reason='bitsandbytes is not supported on this GPU type.')
@pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"),
reason='bitsandbytes is not supported on this GPU type.')
def test_load_bnb_model(vllm_runner) -> None:
with vllm_runner('huggyllama/llama-7b',
quantization='bitsandbytes',