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

Revert "[CI/Build] Add is_quant_method_supported to control quantization test configurations" (#5463)

Browse Source
This commit is contained in:
Simon Mo
2024-06-12 12:03:24 -05:00
committed by GitHub
parent 3dd6853bc8
commit e3c12bf6d2
8 changed files with 68 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tests/quantization/test_bitsandbytes.py
View File

@ -5,12 +5,16 @@ 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(not is_quant_method_supported("bitsandbytes"),
reason='bitsandbytes is not supported on this GPU type.')
@pytest.mark.skipif(
capability < QUANTIZATION_METHODS['bitsandbytes'].get_min_capability(),
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',

10
tests/quantization/test_fp8.py
View File

@ -5,12 +5,16 @@ Run `pytest tests/quantization/test_fp8.py --forked`.
import pytest
import torch
from tests.quantization.utils import is_quant_method_supported
from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
from vllm.model_executor.layers.quantization.fp8 import Fp8LinearMethod
capability = torch.cuda.get_device_capability()
capability = capability[0] * 10 + capability[1]
@pytest.mark.skipif(not is_quant_method_supported("fp8"),
reason="FP8 is not supported on this GPU type.")
@pytest.mark.skipif(
capability < QUANTIZATION_METHODS["fp8"].get_min_capability(),
reason="FP8 is not supported on this GPU type.")
def test_load_fp16_model(vllm_runner) -> None:
with vllm_runner("facebook/opt-125m", quantization="fp8") as llm:

14
tests/quantization/utils.py
View File

@ -1,14 +0,0 @@
import torch
from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
def is_quant_method_supported(quant_method: str) -> bool:
# Currently, all quantization methods require Nvidia or AMD GPUs
if not torch.cuda.is_available():
return False
capability = torch.cuda.get_device_capability()
capability = capability[0] * 10 + capability[1]
return (capability <
QUANTIZATION_METHODS[quant_method].get_min_capability())