xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 01:35:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc] Clean up mark to fork process in BNB tests (#20692)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
Isotr0py 2025-07-10 21:59:40 +08:00 committed by GitHub
parent 1a4f35e2ea
commit 77f77a951e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 11 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
tests/quantization/test_bitsandbytes.py → tests/models/quantization/test_bitsandbytes.py
View File

@ -13,8 +13,8 @@ from transformers import BitsAndBytesConfig
from tests.quantization.utils import is_quant_method_supported from tests.quantization.utils import is_quant_method_supported
from ..models.utils import check_embeddings_close from ...utils import compare_two_settings, multi_gpu_test
from ..utils import compare_two_settings, create_new_process_for_each_test from ..utils import check_embeddings_close
models_4bit_to_test = [ models_4bit_to_test = [
("facebook/opt-125m", "quantize opt model inflight"), ("facebook/opt-125m", "quantize opt model inflight"),
@ -42,7 +42,6 @@ models_pre_quant_8bit_to_test = [
@pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"), @pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"),
reason='bitsandbytes is not supported on this GPU type.') reason='bitsandbytes is not supported on this GPU type.')
@pytest.mark.parametrize("model_name, description", models_4bit_to_test) @pytest.mark.parametrize("model_name, description", models_4bit_to_test)
@create_new_process_for_each_test()
def test_load_4bit_bnb_model(hf_runner, vllm_runner, example_prompts, def test_load_4bit_bnb_model(hf_runner, vllm_runner, example_prompts,
model_name, description) -> None: model_name, description) -> None:
@ -56,7 +55,6 @@ def test_load_4bit_bnb_model(hf_runner, vllm_runner, example_prompts,
reason='bitsandbytes is not supported on this GPU type.') reason='bitsandbytes is not supported on this GPU type.')
@pytest.mark.parametrize("model_name, description", @pytest.mark.parametrize("model_name, description",
models_pre_qaunt_4bit_to_test) models_pre_qaunt_4bit_to_test)
@create_new_process_for_each_test()
def test_load_pre_quant_4bit_bnb_model(hf_runner, vllm_runner, example_prompts, def test_load_pre_quant_4bit_bnb_model(hf_runner, vllm_runner, example_prompts,
model_name, description) -> None: model_name, description) -> None:
@ -68,7 +66,6 @@ def test_load_pre_quant_4bit_bnb_model(hf_runner, vllm_runner, example_prompts,
reason='bitsandbytes is not supported on this GPU type.') reason='bitsandbytes is not supported on this GPU type.')
@pytest.mark.parametrize("model_name, description", @pytest.mark.parametrize("model_name, description",
models_pre_quant_8bit_to_test) models_pre_quant_8bit_to_test)
@create_new_process_for_each_test()
def test_load_8bit_bnb_model(hf_runner, vllm_runner, example_prompts, def test_load_8bit_bnb_model(hf_runner, vllm_runner, example_prompts,
model_name, description) -> None: model_name, description) -> None:
@ -76,12 +73,10 @@ def test_load_8bit_bnb_model(hf_runner, vllm_runner, example_prompts,
model_name, True) model_name, True)
@pytest.mark.skipif(torch.cuda.device_count() < 2,
reason='Test requires at least 2 GPUs.')
@pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"), @pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"),
reason='bitsandbytes is not supported on this GPU type.') reason='bitsandbytes is not supported on this GPU type.')
@pytest.mark.parametrize("model_name, description", models_4bit_to_test) @pytest.mark.parametrize("model_name, description", models_4bit_to_test)
@create_new_process_for_each_test() @multi_gpu_test(num_gpus=2)
def test_load_tp_4bit_bnb_model(hf_runner, vllm_runner, example_prompts, def test_load_tp_4bit_bnb_model(hf_runner, vllm_runner, example_prompts,
model_name, description) -> None: model_name, description) -> None:
@ -96,12 +91,10 @@ def test_load_tp_4bit_bnb_model(hf_runner, vllm_runner, example_prompts,
vllm_tp_size=2) vllm_tp_size=2)
@pytest.mark.skipif(torch.cuda.device_count() < 2,
reason='Test requires at least 2 GPUs.')
@pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"), @pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"),
reason='bitsandbytes is not supported on this GPU type.') reason='bitsandbytes is not supported on this GPU type.')
@pytest.mark.parametrize("model_name, description", models_4bit_to_test) @pytest.mark.parametrize("model_name, description", models_4bit_to_test)
@create_new_process_for_each_test() @multi_gpu_test(num_gpus=2)
def test_load_pp_4bit_bnb_model(model_name, description) -> None: def test_load_pp_4bit_bnb_model(model_name, description) -> None:
common_args = [ common_args = [
"--disable-log-stats", "--disable-log-stats",
@ -127,7 +120,6 @@ def test_load_pp_4bit_bnb_model(model_name, description) -> None:
@pytest.mark.parametrize("model_name, description", @pytest.mark.parametrize("model_name, description",
models_4bit_to_embedding_test) models_4bit_to_embedding_test)
@pytest.mark.parametrize("dtype", ["half"]) @pytest.mark.parametrize("dtype", ["half"])
@create_new_process_for_each_test()
def test_4bit_bnb_embedding_model( def test_4bit_bnb_embedding_model(
model_name, model_name,
description, description,
@ -146,6 +138,13 @@ def test_4bit_bnb_embedding_model(
example_prompts = [str(s).strip() for s in example_prompts] example_prompts = [str(s).strip() for s in example_prompts]
# Inflight 4bit quantization # Inflight 4bit quantization
with vllm_runner(model_name,
task="embed",
dtype=dtype,
gpu_memory_utilization=0.5,
quantization="bitsandbytes") as vllm_model:
vllm_outputs = vllm_model.embed(example_prompts)
hf_model_kwargs = dict(quantization_config=BitsAndBytesConfig( hf_model_kwargs = dict(quantization_config=BitsAndBytesConfig(
load_in_4bit=True)) load_in_4bit=True))
with hf_runner( with hf_runner(
@ -156,12 +155,6 @@ def test_4bit_bnb_embedding_model(
) as hf_model: ) as hf_model:
hf_outputs = hf_model.encode(example_prompts) hf_outputs = hf_model.encode(example_prompts)
with vllm_runner(model_name,
task="embed",
dtype=dtype,
gpu_memory_utilization=0.5,
quantization="bitsandbytes") as vllm_model:
vllm_outputs = vllm_model.embed(example_prompts)
check_embeddings_close( check_embeddings_close(
embeddings_0_lst=hf_outputs, embeddings_0_lst=hf_outputs,
embeddings_1_lst=vllm_outputs, embeddings_1_lst=vllm_outputs,