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

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

129
tests/models/test_transformers.py
View File

@@ -1,6 +1,7 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Test the functionality of the Transformers backend."""
from typing import Any, Optional, Union
import pytest
@@ -60,14 +61,16 @@ def check_implementation(
@pytest.mark.skipif(
current_platform.is_rocm(),
reason="Llama-3.2-1B-Instruct, Ilama-3.2-1B produce memory access fault.")
reason="Llama-3.2-1B-Instruct, Ilama-3.2-1B produce memory access fault.",
)
@pytest.mark.parametrize(
"model,model_impl",
[
("meta-llama/Llama-3.2-1B-Instruct", "transformers"),
("hmellor/Ilama-3.2-1B", "auto"), # CUSTOM CODE
("allenai/OLMoE-1B-7B-0924", "transformers"), # MoE
]) # trust_remote_code=True by default
],
) # trust_remote_code=True by default
def test_models(
hf_runner: type[HfRunner],
vllm_runner: type[VllmRunner],
@@ -77,29 +80,32 @@ def test_models(
) -> None:
import transformers
from packaging.version import Version
installed = Version(transformers.__version__)
required = Version("4.57.0.dev0")
if model == "allenai/OLMoE-1B-7B-0924" and installed < required:
pytest.skip("MoE models with the Transformers backend require "
f"transformers>={required}, but got {installed}")
pytest.skip(
"MoE models with the Transformers backend require "
f"transformers>={required}, but got {installed}"
)
check_implementation(hf_runner,
vllm_runner,
example_prompts,
model,
model_impl=model_impl)
check_implementation(
hf_runner, vllm_runner, example_prompts, model, model_impl=model_impl
)
def test_hybrid_attention(vllm_runner: type[VllmRunner]) -> None:
prompts, _, _ = prep_prompts(4, (800, 801))
kwargs_ref = {"max_model_len": 8192, "enforce_eager": True}
kwargs_test = {"model_impl": "transformers", **kwargs_ref}
check_implementation(vllm_runner,
vllm_runner,
prompts,
model="hmellor/tiny-random-Gemma2ForCausalLM",
kwargs_ref=kwargs_ref,
kwargs_test=kwargs_test)
check_implementation(
vllm_runner,
vllm_runner,
prompts,
model="hmellor/tiny-random-Gemma2ForCausalLM",
kwargs_ref=kwargs_ref,
kwargs_test=kwargs_test,
)
@multi_gpu_test(num_gpus=2)
@@ -109,23 +115,28 @@ def test_distributed(
example_prompts,
):
kwargs = {"model_impl": "transformers", "tensor_parallel_size": 2}
check_implementation(hf_runner,
vllm_runner,
example_prompts,
"meta-llama/Llama-3.2-1B-Instruct",
kwargs_test=kwargs)
@pytest.mark.parametrize("model, quantization_kwargs", [
("TheBloke/TinyLlama-1.1B-Chat-v0.3-AWQ", {}),
("TheBloke/TinyLlama-1.1B-Chat-v0.3-GPTQ", {}),
(
check_implementation(
hf_runner,
vllm_runner,
example_prompts,
"meta-llama/Llama-3.2-1B-Instruct",
{
"quantization": "bitsandbytes",
},
),
])
kwargs_test=kwargs,
)
@pytest.mark.parametrize(
"model, quantization_kwargs",
[
("TheBloke/TinyLlama-1.1B-Chat-v0.3-AWQ", {}),
("TheBloke/TinyLlama-1.1B-Chat-v0.3-GPTQ", {}),
(
"meta-llama/Llama-3.2-1B-Instruct",
{
"quantization": "bitsandbytes",
},
),
],
)
@pytest.mark.parametrize("max_tokens", [32])
@pytest.mark.parametrize("num_logprobs", [5])
def test_quantization(
@@ -136,27 +147,34 @@ def test_quantization(
max_tokens: int,
num_logprobs: int,
) -> None:
if (current_platform.is_rocm()
and quantization_kwargs.get("quantization", "") == "bitsandbytes"):
pytest.skip(
"bitsandbytes quantization is currently not supported in rocm.")
if (
current_platform.is_rocm()
and quantization_kwargs.get("quantization", "") == "bitsandbytes"
):
pytest.skip("bitsandbytes quantization is currently not supported in rocm.")
with vllm_runner(
model, model_impl="auto", enforce_eager=True,
**quantization_kwargs) as vllm_model: # type: ignore[arg-type]
model,
model_impl="auto",
enforce_eager=True,
**quantization_kwargs, # type: ignore[arg-type]
) as vllm_model:
vllm_outputs = vllm_model.generate_greedy_logprobs(
example_prompts, max_tokens=max_tokens, num_logprobs=num_logprobs)
example_prompts, max_tokens=max_tokens, num_logprobs=num_logprobs
)
with vllm_runner(
model,
model_impl="transformers",
enforce_eager=True,
**quantization_kwargs) as vllm_model: # type: ignore[arg-type]
model,
model_impl="transformers",
enforce_eager=True,
**quantization_kwargs, # type: ignore[arg-type]
) as vllm_model:
model_config = vllm_model.llm.llm_engine.model_config
assert model_config.using_transformers_backend()
transformers_outputs = vllm_model.generate_greedy_logprobs(
example_prompts, max_tokens=max_tokens, num_logprobs=num_logprobs)
example_prompts, max_tokens=max_tokens, num_logprobs=num_logprobs
)
check_logprobs_close(
outputs_0_lst=transformers_outputs,
@@ -172,22 +190,24 @@ def test_quantization(
# Layers live in `layers`
"Qwen/Qwen3-Embedding-0.6B",
# Layers live in `model.layers`
"meta-llama/Llama-3.2-1B-Instruct"
"meta-llama/Llama-3.2-1B-Instruct",
],
)
def test_embed_loading(vllm_runner, model):
with vllm_runner(model,
max_model_len=1024,
enforce_eager=True,
runner="pooling",
model_impl="transformers") as model_test:
with vllm_runner(
model,
max_model_len=1024,
enforce_eager=True,
runner="pooling",
model_impl="transformers",
) as model_test:
model_config = model_test.llm.llm_engine.model_config
assert model_config.using_transformers_backend()
@pytest.mark.parametrize(
"arch",
["TransformersEmbeddingModel", "TransformersForSequenceClassification"])
"arch", ["TransformersEmbeddingModel", "TransformersForSequenceClassification"]
)
def test_pooling(hf_runner, vllm_runner, example_prompts, arch):
model = get_model(arch)
@@ -202,6 +222,7 @@ def test_pooling(hf_runner, vllm_runner, example_prompts, arch):
hf_kwargs["is_sentence_transformer"] = True
elif arch == "TransformersForSequenceClassification":
from transformers import AutoModelForSequenceClassification
hf_kwargs["auto_cls"] = AutoModelForSequenceClassification
# The example_prompts has ending "\n", for example:
@@ -212,8 +233,10 @@ def test_pooling(hf_runner, vllm_runner, example_prompts, arch):
# So we need to strip the input texts to avoid test failing.
example_prompts = [str(s).strip() for s in example_prompts]
with (vllm_runner(model, **vllm_kwargs) as
vllm_model, hf_runner(model, **hf_kwargs) as hf_model):
with (
vllm_runner(model, **vllm_kwargs) as vllm_model,
hf_runner(model, **hf_kwargs) as hf_model,
):
model_config = vllm_model.llm.llm_engine.model_config
assert model_config.using_transformers_backend()