[Misc] Clean up test docstrings and names (#17521)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

.buildkite/test-pipeline.yaml

@@ -395,10 +395,8 @@ steps:
   - csrc/
   - vllm/model_executor/layers/quantization
   - tests/quantization
-  - tests/models/quantization
   commands:
   - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization
-  - pytest -v -s models/quantization
 
 - label: LM Eval Small Models # 53min
   working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
@@ -509,6 +507,14 @@ steps:
     - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
     - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
 
+- label: Quantized Models Test
+  #mirror_hardwares: [amd]
+  source_file_dependencies:
+  - vllm/model_executor/layers/quantization
+  - tests/models/quantization
+  commands:
+    - pytest -v -s models/quantization
+
 # This test is used only in PR development phase to test individual models and should never run on main
 - label: Custom Models Test
   mirror_hardwares: [amd]

tests/models/language/generation/test_models.py → tests/models/language/generation/test_common.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the outputs of HF and vLLM when using greedy sampling.
-
-Run `pytest tests/models/test_models.py`.
-"""
-
 import pytest
 import torch
 

tests/models/language/generation/test_granite.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the outputs of HF and vLLM for Granite models using greedy sampling.
-
-Run `pytest tests/models/test_granite.py`.
-"""
 import pytest
 
 from ...utils import check_logprobs_close

tests/models/language/generation/test_mistral.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the outputs of HF and vLLM for Mistral models using greedy sampling.
-
-Run `pytest tests/models/test_mistral.py`.
-"""
 import copy
 import json
 

tests/models/language/generation/test_phimoe.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the outputs of HF and vLLM for moe models using greedy sampling.
-
-Run `pytest tests/models/test_phimoe.py`.
-"""
 import pytest
 import torch
 

tests/models/language/pooling/test_cls_models.py → tests/models/language/pooling/test_classification.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the classification outputs of HF and vLLM models.
-
-Run `pytest tests/models/test_cls_models.py`.
-"""
 import pytest
 import torch
 from transformers import AutoModelForSequenceClassification
@@ -19,7 +15,7 @@ from vllm.platforms import current_platform
 )
 @pytest.mark.parametrize("dtype",
                          ["half"] if current_platform.is_rocm() else ["float"])
-def test_classification_models(
+def test_models(
     hf_runner,
     vllm_runner,
     example_prompts,

tests/models/language/pooling/test_embedding.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the embedding outputs of HF and vLLM models.
-
-Run `pytest tests/models/embedding/language/test_embedding.py`.
-"""
 import pytest
 
 from vllm.config import PoolerConfig

tests/models/language/pooling/test_jina.py

 # SPDX-License-Identifier: Apache-2.0
-# ruff: noqa: E501
-"""Compare the scoring outputs of HF and vLLM models.
-
-Run `pytest tests/models/embedding/language/test_jina.py`.
-"""
 import math
 
 import pytest
@@ -22,9 +17,9 @@ TEXTS_2 = [
     "Organic skincare for sensitive skin with aloe vera and chamomile.",
     "New makeup trends focus on bold colors and innovative techniques",
     "Bio-Hautpflege für empfindliche Haut mit Aloe Vera und Kamille",
-    "Neue Make-up-Trends setzen auf kräftige Farben und innovative Techniken",
-    "Cuidado de la piel orgánico para piel sensible con aloe vera y manzanilla",
-    "Las nuevas tendencias de maquillaje se centran en colores vivos y técnicas innovadoras",
+    "Neue Make-up-Trends setzen auf kräftige Farben und innovative Techniken",  # noqa: E501
+    "Cuidado de la piel orgánico para piel sensible con aloe vera y manzanilla",  # noqa: E501
+    "Las nuevas tendencias de maquillaje se centran en colores vivos y técnicas innovadoras",  # noqa: E501
     "针对敏感肌专门设计的天然有机护肤产品",
     "新的化妆趋势注重鲜艳的颜色和创新的技巧",
     "敏感肌のために特別に設計された天然有機スキンケア製品",

tests/models/language/pooling/test_scoring.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the scoring outputs of HF and vLLM models.
-
-Run `pytest tests/models/embedding/language/test_scoring.py`.
-"""
 import math
 
 import pytest
 import torch
 import torch.nn.functional as F
 
-MODELS = [
+CROSS_ENCODER_MODELS = [
     "cross-encoder/ms-marco-MiniLM-L-6-v2",  # Bert
     "BAAI/bge-reranker-v2-m3",  # Roberta
 ]
@@ -28,21 +24,21 @@ TEXTS_2 = [
     "The capital of Germany is Berlin.",
 ]
 
+DTYPE = "half"
+
 
-@pytest.fixture(scope="module", params=MODELS)
+@pytest.fixture(scope="module", params=CROSS_ENCODER_MODELS)
 def model_name(request):
     yield request.param
 
 
-@pytest.mark.parametrize("dtype", ["half"])
-def test_llm_1_to_1(vllm_runner, hf_runner, model_name, dtype: str):
-
+def test_cross_encoder_1_to_1(vllm_runner, hf_runner, model_name):
     text_pair = [TEXTS_1[0], TEXTS_2[0]]
 
-    with hf_runner(model_name, dtype=dtype, is_cross_encoder=True) as hf_model:
+    with hf_runner(model_name, dtype=DTYPE, is_cross_encoder=True) as hf_model:
         hf_outputs = hf_model.predict([text_pair]).tolist()
 
-    with vllm_runner(model_name, task="score", dtype=dtype,
+    with vllm_runner(model_name, task="score", dtype=DTYPE,
                      max_model_len=None) as vllm_model:
         vllm_outputs = vllm_model.score(text_pair[0], text_pair[1])
 
@@ -52,18 +48,16 @@ def test_llm_1_to_1(vllm_runner, hf_runner, model_name, dtype: str):
     assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
 
 
-@pytest.mark.parametrize("dtype", ["half"])
-def test_llm_1_to_N(vllm_runner, hf_runner, model_name, dtype: str):
-
+def test_cross_encoder_1_to_N(vllm_runner, hf_runner, model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[0], TEXTS_2[1]],
     ]
 
-    with hf_runner(model_name, dtype=dtype, is_cross_encoder=True) as hf_model:
+    with hf_runner(model_name, dtype=DTYPE, is_cross_encoder=True) as hf_model:
         hf_outputs = hf_model.predict(text_pairs).tolist()
 
-    with vllm_runner(model_name, task="score", dtype=dtype,
+    with vllm_runner(model_name, task="score", dtype=DTYPE,
                      max_model_len=None) as vllm_model:
         vllm_outputs = vllm_model.score(TEXTS_1[0], TEXTS_2)
 
@@ -74,18 +68,16 @@ def test_llm_1_to_N(vllm_runner, hf_runner, model_name, dtype: str):
     assert math.isclose(hf_outputs[1], vllm_outputs[1], rel_tol=0.01)
 
 
-@pytest.mark.parametrize("dtype", ["half"])
-def test_llm_N_to_N(vllm_runner, hf_runner, model_name, dtype: str):
-
+def test_cross_encoder_N_to_N(vllm_runner, hf_runner, model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[1], TEXTS_2[1]],
     ]
 
-    with hf_runner(model_name, dtype=dtype, is_cross_encoder=True) as hf_model:
+    with hf_runner(model_name, dtype=DTYPE, is_cross_encoder=True) as hf_model:
         hf_outputs = hf_model.predict(text_pairs).tolist()
 
-    with vllm_runner(model_name, task="score", dtype=dtype,
+    with vllm_runner(model_name, task="score", dtype=DTYPE,
                      max_model_len=None) as vllm_model:
         vllm_outputs = vllm_model.score(TEXTS_1, TEXTS_2)
 
@@ -101,13 +93,10 @@ def emb_model_name(request):
     yield request.param
 
 
-@pytest.mark.parametrize("dtype", ["half"])
-def test_llm_1_to_1_embedding(vllm_runner, hf_runner, emb_model_name,
-                              dtype: str):
-
+def test_embedding_1_to_1(vllm_runner, hf_runner, emb_model_name):
     text_pair = [TEXTS_1[0], TEXTS_2[0]]
 
-    with hf_runner(emb_model_name, dtype=dtype,
+    with hf_runner(emb_model_name, dtype=DTYPE,
                    is_sentence_transformer=True) as hf_model:
         hf_embeddings = hf_model.encode(text_pair)
         hf_outputs = [
@@ -116,7 +105,7 @@ def test_llm_1_to_1_embedding(vllm_runner, hf_runner, emb_model_name,
 
     with vllm_runner(emb_model_name,
                      task="embed",
-                     dtype=dtype,
+                     dtype=DTYPE,
                      max_model_len=None) as vllm_model:
         vllm_outputs = vllm_model.score(text_pair[0], text_pair[1])
 
@@ -126,16 +115,13 @@ def test_llm_1_to_1_embedding(vllm_runner, hf_runner, emb_model_name,
     assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
 
 
-@pytest.mark.parametrize("dtype", ["half"])
-def test_llm_1_to_N_embedding(vllm_runner, hf_runner, emb_model_name,
-                              dtype: str):
-
+def test_embedding_1_to_N(vllm_runner, hf_runner, emb_model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[0], TEXTS_2[1]],
     ]
 
-    with hf_runner(emb_model_name, dtype=dtype,
+    with hf_runner(emb_model_name, dtype=DTYPE,
                    is_sentence_transformer=True) as hf_model:
         hf_embeddings = [
             hf_model.encode(text_pair) for text_pair in text_pairs
@@ -147,7 +133,7 @@ def test_llm_1_to_N_embedding(vllm_runner, hf_runner, emb_model_name,
 
     with vllm_runner(emb_model_name,
                      task="embed",
-                     dtype=dtype,
+                     dtype=DTYPE,
                      max_model_len=None) as vllm_model:
         vllm_outputs = vllm_model.score(TEXTS_1[0], TEXTS_2)
 
@@ -158,16 +144,13 @@ def test_llm_1_to_N_embedding(vllm_runner, hf_runner, emb_model_name,
     assert math.isclose(hf_outputs[1], vllm_outputs[1], rel_tol=0.01)
 
 
-@pytest.mark.parametrize("dtype", ["half"])
-def test_llm_N_to_N_embedding(vllm_runner, hf_runner, emb_model_name,
-                              dtype: str):
-
+def test_embedding_N_to_N(vllm_runner, hf_runner, emb_model_name):
     text_pairs = [
         [TEXTS_1[0], TEXTS_2[0]],
         [TEXTS_1[1], TEXTS_2[1]],
     ]
 
-    with hf_runner(emb_model_name, dtype=dtype,
+    with hf_runner(emb_model_name, dtype=DTYPE,
                    is_sentence_transformer=True) as hf_model:
         hf_embeddings = [
             hf_model.encode(text_pair) for text_pair in text_pairs
@@ -179,7 +162,7 @@ def test_llm_N_to_N_embedding(vllm_runner, hf_runner, emb_model_name,
 
     with vllm_runner(emb_model_name,
                      task="embed",
-                     dtype=dtype,
+                     dtype=DTYPE,
                      max_model_len=None) as vllm_model:
         vllm_outputs = vllm_model.score(TEXTS_1, TEXTS_2)
 

tests/models/language/pooling/test_snowflake_arctic_embed.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the embedding outputs of HF and vLLM models.
-
-Run `pytest tests/models/embedding/language/test_snowflake_arctic_embed.py`.
-"""
 import pytest
 
 from ...utils import EmbedModelInfo, check_embeddings_close

tests/models/language/pooling/test_truncation_control.py

@@ -5,18 +5,18 @@ MODEL_NAME = "sentence-transformers/all-MiniLM-L12-v2"
 max_model_len = 128
 
 input_str = """Immerse yourself in the enchanting chronicle of calculus, a 
-    mathematical domain that has radically transformed our comprehension of 
-    change and motion. Despite its roots in ancient civilizations, the 
-    formal birth of calculus predominantly occurred in the 17th century, 
-    primarily under the influential guidance of Sir Isaac Newton and Gottfried 
-    Wilhelm Leibniz. The earliest traces of calculus concepts are found in 
-    ancient Greek mathematics,most notably in the works of Eudoxus and 
-    Archimedes, around 300 BCE. They utilized the 'method of exhaustion'—a 
-    technique for computing areas and volumes through the use of finite sums. 
-    This methodology laid crucial foundational work for integral calculus. 
-    In the 17th century, both Newton and Leibniz independently pioneered 
-    calculus, each contributing unique perspectives that would shape this new 
-    field."""
+mathematical domain that has radically transformed our comprehension of 
+change and motion. Despite its roots in ancient civilizations, the 
+formal birth of calculus predominantly occurred in the 17th century, 
+primarily under the influential guidance of Sir Isaac Newton and Gottfried 
+Wilhelm Leibniz. The earliest traces of calculus concepts are found in 
+ancient Greek mathematics,most notably in the works of Eudoxus and 
+Archimedes, around 300 BCE. They utilized the 'method of exhaustion'—a 
+technique for computing areas and volumes through the use of finite sums. 
+This methodology laid crucial foundational work for integral calculus. 
+In the 17th century, both Newton and Leibniz independently pioneered 
+calculus, each contributing unique perspectives that would shape this new 
+field."""
 
 
 def test_smaller_truncation_size(vllm_runner,

tests/models/multimodal/generation/test_pixtral.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the outputs of HF and vLLM for Mistral models using greedy sampling.
-
-Run `pytest tests/models/test_mistral.py`.
-"""
 import json
 from dataclasses import asdict
 from typing import TYPE_CHECKING, Any, Optional

tests/models/multimodal/generation/test_whisper.py

@@ -119,10 +119,10 @@ def run_test(
         assert output.outputs[0].text == expected
 
 
-@create_new_process_for_each_test("spawn")
 @pytest.mark.core_model
 @pytest.mark.parametrize(
     "model", ["openai/whisper-small", "openai/whisper-large-v3-turbo"])
+@create_new_process_for_each_test()
 def test_models(vllm_runner, model) -> None:
     run_test(
         vllm_runner,
@@ -131,11 +131,11 @@ def test_models(vllm_runner, model) -> None:
     )
 
 
-@create_new_process_for_each_test("spawn")
 @multi_gpu_test(num_gpus=2)
 @pytest.mark.core_model
 @pytest.mark.parametrize("model", ["openai/whisper-large-v3-turbo"])
 @pytest.mark.parametrize("distributed_executor_backend", ["ray", "mp"])
+@create_new_process_for_each_test()
 def test_models_distributed(
     vllm_runner,
     model,

tests/models/quantization/test_aqlm.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compare the outputs of a AQLM model between vLLM and HF Transformers
-
-Run `pytest tests/models/test_aqlm.py`.
-"""
-
 import pytest
 
 from tests.quantization.utils import is_quant_method_supported

tests/models/quantization/test_bitblas.py

@@ -8,8 +8,6 @@ bitblas/GPTQ models are in the top 3 selections of each other.
 Note: bitblas internally uses locks to synchronize the threads. This can
 result in very slight nondeterminism for bitblas. As a result, we re-run the 
 test up to 3 times to see if we pass.
-
-Run `pytest tests/models/test_bitblas.py`.
 """
 from dataclasses import dataclass
 

tests/models/quantization/test_gptq_bitblas.py

@@ -8,8 +8,6 @@ bitblas/GPTQ models are in the top 3 selections of each other.
 Note: bitblas internally uses locks to synchronize the threads. This can
 result in very slight nondeterminism for bitblas. As a result, we re-run the 
 test up to 3 times to see if we pass.
-
-Run `pytest tests/models/test_bitblas.py`.
 """
 from dataclasses import dataclass
 

tests/models/quantization/test_gptq_marlin.py

 # SPDX-License-Identifier: Apache-2.0
-"""Compares the outputs of gptq vs gptq_marlin 
+"""Compares the outputs of gptq vs gptq_marlin.
+
 Note: GPTQ and Marlin do not have bitwise correctness.
 As a result, in this test, we just confirm that the top selected tokens of the
 Marlin/GPTQ models are in the top 5 selections of each other.
 Note: Marlin internally uses locks to synchronize the threads. This can
 result in very slight nondeterminism for Marlin. As a result, we re-run the test
 up to 3 times to see if we pass.
-
-Run `pytest tests/models/test_gptq_marlin.py`.
 """
 import os
 

tests/models/quantization/test_gptq_marlin_24.py

@@ -4,8 +4,6 @@
 Note: GPTQ and Marlin_24 do not have bitwise correctness.
 As a result, in this test, we just confirm that the top selected tokens of the
 Marlin/GPTQ models are in the top 3 selections of each other.
-
-Run `pytest tests/models/test_marlin_24.py`.
 """
 from dataclasses import dataclass
 

tests/models/test_transformers.py

 # SPDX-License-Identifier: Apache-2.0
-"""Test the functionality of the Transformers backend.
-
-Run `pytest tests/models/test_transformers.py`.
-"""
+"""Test the functionality of the Transformers backend."""
 import pytest
 
 from ..conftest import HfRunner, VllmRunner