384dc7f77b · tduncan/Nvidia-Optimization-Stack-VLLM

.buildkite/scripts/hardware_ci/run-amd-test.sh

+4 -4

@@ -333,15 +333,15 @@ apply_rocm_test_overrides() {

333 # --- Entrypoint ignores ---

334 if [[ $cmds == *" entrypoints/openai "* ]]; then

335 cmds=${cmds//" entrypoints/openai "/" entrypoints/openai \

336 - --ignore=entrypoints/openai/test_audio.py \

337 - --ignore=entrypoints/openai/test_shutdown.py \

336 + --ignore=entrypoints/openai/chat_completion/test_audio.py \

337 + --ignore=entrypoints/openai/completion/test_shutdown.py \

338 --ignore=entrypoints/openai/test_completion.py \

339 --ignore=entrypoints/openai/test_models.py \

340 --ignore=entrypoints/openai/test_lora_adapters.py \

341 --ignore=entrypoints/openai/test_return_tokens_as_ids.py \

342 - --ignore=entrypoints/openai/test_root_path.py \

342 + --ignore=entrypoints/openai/chat_completion/test_root_path.py \

343 --ignore=entrypoints/openai/test_tokenization.py \

344 - --ignore=entrypoints/openai/test_prompt_validation.py "}

344 + --ignore=entrypoints/openai/completion/test_prompt_validation.py "}

345 fi

346

347 if [[ $cmds == *" entrypoints/llm "* ]]; then

.buildkite/test-amd.yaml

+12 -12

@@ -162,7 +162,7 @@ steps:

162 - tests/entrypoints/test_chat_utils

163 commands:

164 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

165 - - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

165 + - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/chat_completion/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/chat_completion/test_oot_registration.py --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

166 - pytest -v -s entrypoints/test_chat_utils.py

167

168 - label: Entrypoints Integration Test (API Server 2)

@@ -674,12 +674,12 @@ steps:

674 - vllm/config/model.py

675 - vllm/model_executor

676 - tests/model_executor

677 - - tests/entrypoints/openai/test_tensorizer_entrypoint.py

677 + - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py

678 commands:

679 - apt-get update && apt-get install -y curl libsodium23

680 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

681 - pytest -v -s model_executor

682 - - pytest -v -s entrypoints/openai/test_tensorizer_entrypoint.py

682 + - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py

683

684 - label: Benchmarks # 11min

685 timeout_in_minutes: 20

@@ -1143,7 +1143,7 @@ steps:

1143 - pytest -v -s plugins_tests/test_scheduler_plugins.py

1144 - pip install -e ./plugins/vllm_add_dummy_model

1145 - pytest -v -s distributed/test_distributed_oot.py

1146 - - pytest -v -s entrypoints/openai/test_oot_registration.py

1146 + - pytest -v -s entrypoints/openai/chat_completion/test_oot_registration.py

1147 - pytest -v -s models/test_oot_registration.py

1148 - pytest -v -s plugins/lora_resolvers

1149

@@ -1502,7 +1502,7 @@ steps:

1502 - tests/entrypoints/test_chat_utils

1503 commands:

1504 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

1505 - - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

1505 + - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/chat_completion/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/chat_completion/test_oot_registration.py --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

1506 - pytest -v -s entrypoints/test_chat_utils.py

1507

1508 - label: Entrypoints Integration Test (API Server 2)

@@ -2133,12 +2133,12 @@ steps:

2133 - vllm/config/model.py

2134 - vllm/model_executor

2135 - tests/model_executor

2136 - - tests/entrypoints/openai/test_tensorizer_entrypoint.py

2136 + - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py

2137 commands:

2138 - apt-get update && apt-get install -y curl libsodium23

2139 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

2140 - pytest -v -s model_executor

2141 - - pytest -v -s entrypoints/openai/test_tensorizer_entrypoint.py

2141 + - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py

2142

2143 - label: Benchmarks # 11min

2144 timeout_in_minutes: 20

@@ -2735,7 +2735,7 @@ steps:

2735 - pytest -v -s plugins_tests/test_scheduler_plugins.py

2736 - pip install -e ./plugins/vllm_add_dummy_model

2737 - pytest -v -s distributed/test_distributed_oot.py

2738 - - pytest -v -s entrypoints/openai/test_oot_registration.py # it needs a clean process

2738 + - pytest -v -s entrypoints/openai/chat_completion/test_oot_registration.py # it needs a clean process

2739 - pytest -v -s models/test_oot_registration.py # it needs a clean process

2740 - pytest -v -s plugins/lora_resolvers # unit tests for in-tree lora resolver plugins

2741

@@ -3257,7 +3257,7 @@ steps:

3257 - tests/entrypoints/test_chat_utils

3258 commands:

3259 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

3260 - - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

3260 + - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/chat_completion/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/chat_completion/test_oot_registration.py --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

3261 - pytest -v -s entrypoints/test_chat_utils.py

3262

3263 - label: Entrypoints Integration Test (API Server 2)

@@ -3872,12 +3872,12 @@ steps:

3872 - vllm/config/model.py

3873 - vllm/model_executor

3874 - tests/model_executor

3875 - - tests/entrypoints/openai/test_tensorizer_entrypoint.py

3875 + - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py

3876 commands:

3877 - apt-get update && apt-get install -y curl libsodium23

3878 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

3879 - pytest -v -s model_executor

3880 - - pytest -v -s entrypoints/openai/test_tensorizer_entrypoint.py

3880 + - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py

3881

3882 - label: Benchmarks # 11min

3883 timeout_in_minutes: 20

@@ -4508,7 +4508,7 @@ steps:

4508 - pytest -v -s plugins_tests/test_scheduler_plugins.py

4509 - pip install -e ./plugins/vllm_add_dummy_model

4510 - pytest -v -s distributed/test_distributed_oot.py

4511 - - pytest -v -s entrypoints/openai/test_oot_registration.py # it needs a clean process

4511 + - pytest -v -s entrypoints/openai/chat_completion/test_oot_registration.py # it needs a clean process

4512 - pytest -v -s models/test_oot_registration.py # it needs a clean process

4513 - pytest -v -s plugins/lora_resolvers # unit tests for in-tree lora resolver plugins

4514

.buildkite/test_areas/entrypoints.yaml

+1 -1

@@ -34,7 +34,7 @@ steps:

34 - tests/entrypoints/test_chat_utils

35 commands:

36 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

37 - - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/chat_completion/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

37 + - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/chat_completion/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/chat_completion/test_oot_registration.py --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/responses

38 - pytest -v -s entrypoints/test_chat_utils.py

39 mirror:

40 amd:

.buildkite/test_areas/model_executor.yaml

+2 -2

@@ -9,9 +9,9 @@ steps:

9 - vllm/config/model.py

10 - vllm/model_executor

11 - tests/model_executor

12 - - tests/entrypoints/openai/test_tensorizer_entrypoint.py

12 + - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py

13 commands:

14 - apt-get update && apt-get install -y curl libsodium23

15 - export VLLM_WORKER_MULTIPROC_METHOD=spawn

16 - pytest -v -s model_executor

17 - - pytest -v -s entrypoints/openai/test_tensorizer_entrypoint.py

17 + - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py

.buildkite/test_areas/plugins.yaml

+1 -1

@@ -36,6 +36,6 @@ steps:

36 - pytest -v -s plugins_tests/test_scheduler_plugins.py

37 - pip install -e ./plugins/vllm_add_dummy_model

38 - pytest -v -s distributed/test_distributed_oot.py

39 - - pytest -v -s entrypoints/openai/test_oot_registration.py # it needs a clean process

39 + - pytest -v -s entrypoints/openai/chat_completion/test_oot_registration.py # it needs a clean process

40 - pytest -v -s models/test_oot_registration.py # it needs a clean process

41 - pytest -v -s plugins/lora_resolvers # unit tests for in-tree lora resolver plugins

.github/mergify.yml

+1 -1

@@ -381,7 +381,7 @@ pull_request_rules:

381 - or:

382 - files~=^vllm/model_executor/model_loader/tensorizer.py

383 - files~=^vllm/model_executor/model_loader/tensorizer_loader.py

384 - - files~=^tests/entrypoints/openai/test_tensorizer_entrypoint.py

384 + - files~=^tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py

385 - files~=^tests/model_executor/model_loader/tensorizer_loader/

386 actions:

387 assign:

requirements/rocm-test.txt

+1 -1

@@ -50,7 +50,7 @@ av==16.1.0

50 blobfile==3.0.0

51 # Multi-Modal Models Test

52 decord==0.6.0

53 - # video processing, required by entrypoints/openai/test_video.py

53 + # video processing, required by entrypoints/openai/chat_completion/test_video.py

54 rapidfuzz==3.12.1

55

56 # OpenAI compatibility and testing

tests/distributed/test_distributed_oot.py

+3 -1

@@ -1,7 +1,9 @@

1 # SPDX-License-Identifier: Apache-2.0

2 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project

3

4 -from ..entrypoints.openai.test_oot_registration import run_and_test_dummy_opt_api_server

4 +from tests.entrypoints.openai.chat_completion.test_oot_registration import (

5 + run_and_test_dummy_opt_api_server,

6 +)

7

8

9 def test_distributed_oot(dummy_opt_path: str):

tests/entrypoints/llm/test_chat.py

+1 -2

@@ -4,12 +4,11 @@ import weakref

4

5 import pytest

6

7 +from tests.entrypoints.openai.chat_completion.test_vision import TEST_IMAGE_ASSETS

8 from vllm import LLM

9 from vllm.distributed import cleanup_dist_env_and_memory

10 from vllm.sampling_params import SamplingParams

11

11 -from ..openai.test_vision import TEST_IMAGE_ASSETS

12 -

12

13 @pytest.fixture(scope="function")

14 def text_llm():

tests/entrypoints/llm/test_mm_cache_stats.py

+1 -2

@@ -6,13 +6,12 @@ import logging

6 import pytest

7 import regex as re

8

9 +from tests.entrypoints.openai.chat_completion.test_vision import TEST_IMAGE_ASSETS

10 from vllm import LLM

11 from vllm.entrypoints.chat_utils import ChatCompletionMessageParam

12 from vllm.v1.metrics import loggers as stat_loggers

13 from vllm.v1.metrics.reader import Counter, Metric

14

14 -from ..openai.test_vision import TEST_IMAGE_ASSETS

15 -

15

16 def _make_messages(image_url: str) -> list[ChatCompletionMessageParam]:

17 return [

tests/entrypoints/openai/chat_completion/test_audio.py renamed

+1 -2

@@ -7,11 +7,10 @@ import openai

7 import pytest

8 import pytest_asyncio

9

10 +from tests.utils import RemoteOpenAIServer

11 from vllm.assets.audio import AudioAsset

12 from vllm.multimodal.utils import encode_audio_base64, encode_audio_url, fetch_audio

13

13 -from ...utils import RemoteOpenAIServer

14 -

14 MODEL_NAME = "fixie-ai/ultravox-v0_5-llama-3_2-1b"

15 TEST_AUDIO_URLS = [

16 AudioAsset("winning_call").url,

tests/entrypoints/openai/chat_completion/test_audio_in_video.py renamed

+2 -2

@@ -8,8 +8,8 @@ import openai

8 import pytest

9 import pytest_asyncio

10

11 -from ...conftest import VideoTestAssets

12 -from ...utils import RemoteOpenAIServer

11 +from tests.conftest import VideoTestAssets

12 +from tests.utils import RemoteOpenAIServer

13

14 MODEL_NAME = "Qwen/Qwen2.5-Omni-3B"

15

tests/entrypoints/openai/chat_completion/test_default_mm_loras.py renamed

+2 -2

@@ -8,8 +8,8 @@ import pytest

8 import pytest_asyncio

9 from huggingface_hub import snapshot_download

10

11 -from ...conftest import AudioTestAssets

12 -from ...utils import RemoteOpenAIServer

11 +from tests.conftest import AudioTestAssets

12 +from tests.utils import RemoteOpenAIServer

13

14 # NOTE - the tests in this module are currently analogous to test_chat, but are

15 # separated to avoid OOM killing due to module-scoped servers, since we

tests/entrypoints/openai/chat_completion/test_oot_registration.py renamed

+1 -1

@@ -1,7 +1,7 @@

1 # SPDX-License-Identifier: Apache-2.0

2 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project

3

4 -from ...utils import VLLM_PATH, RemoteOpenAIServer

4 +from tests.utils import VLLM_PATH, RemoteOpenAIServer

5

6 chatml_jinja_path = VLLM_PATH / "examples/template_chatml.jinja"

7 assert chatml_jinja_path.exists()

tests/entrypoints/openai/chat_completion/test_root_path.py renamed

+1 -1

@@ -8,7 +8,7 @@ from typing import Any, NamedTuple

8 import openai # use the official client for correctness check

9 import pytest

10

11 -from ...utils import RemoteOpenAIServer

11 +from tests.utils import RemoteOpenAIServer

12

13 # # any model with a chat template should work here

14 MODEL_NAME = "Qwen/Qwen2-1.5B-Instruct"

tests/entrypoints/openai/chat_completion/test_video.py renamed

+1 -2

@@ -7,11 +7,10 @@ import openai

7 import pytest

8 import pytest_asyncio

9

10 +from tests.utils import RemoteOpenAIServer

11 from vllm.multimodal.utils import encode_video_url, fetch_video

12 from vllm.platforms import current_platform

13

13 -from ...utils import RemoteOpenAIServer

14 -

14 MODEL_NAME = "llava-hf/llava-onevision-qwen2-0.5b-ov-hf"

15 MAXIMUM_VIDEOS = 3

16

tests/entrypoints/openai/chat_completion/test_vision.py renamed

+1 -2

@@ -8,12 +8,11 @@ import pytest

8 import pytest_asyncio

9 from transformers import AutoProcessor

10

11 +from tests.utils import ROCM_ENV_OVERRIDES, ROCM_EXTRA_ARGS, RemoteOpenAIServer

12 from vllm.multimodal.media import MediaWithBytes

13 from vllm.multimodal.utils import encode_image_url, fetch_image

14 from vllm.platforms import current_platform

15

15 -from ...utils import ROCM_ENV_OVERRIDES, ROCM_EXTRA_ARGS, RemoteOpenAIServer

16 -

16 MODEL_NAME = "microsoft/Phi-3.5-vision-instruct"

17 MAXIMUM_IMAGES = 2

18

tests/entrypoints/openai/chat_completion/test_vision_embeds.py renamed

+1 -2

@@ -8,10 +8,9 @@ import pytest

8 import requests

9 import torch

10

11 +from tests.utils import RemoteOpenAIServer

12 from vllm.utils.serial_utils import tensor2base64

13

13 -from ...utils import RemoteOpenAIServer

14 -

14

15 @pytest.mark.parametrize(

16 "model_name", ["ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL-Sen1Floods11"]

tests/entrypoints/openai/completion/__init__.py

tests/entrypoints/openai/completion/test_completion_error.py renamed

tests/entrypoints/openai/completion/test_completion_with_prompt_embeds.py renamed

+1 -1

@@ -14,7 +14,7 @@ import torch

14 from openai import BadRequestError

15 from transformers import AutoConfig

16

17 -from ...utils import RemoteOpenAIServer

17 +from tests.utils import RemoteOpenAIServer

18

19 # any model with a chat template should work here

20 MODEL_NAME = "facebook/opt-125m"

tests/entrypoints/openai/completion/test_lora_resolvers.py renamed

tests/entrypoints/openai/completion/test_prompt_validation.py renamed

+1 -2

@@ -11,11 +11,10 @@ import pytest

11 import regex as re

12 import torch

13

14 +from tests.utils import RemoteOpenAIServer

15 from vllm.config import ModelConfig

16 from vllm.renderers.embed_utils import safe_load_prompt_embeds

17

17 -from ...utils import RemoteOpenAIServer

18 -

18

19 @pytest.mark.asyncio

20 async def test_empty_prompt():

tests/entrypoints/openai/completion/test_shutdown.py renamed

tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py renamed

+1 -2

@@ -9,6 +9,7 @@ import pytest

9 import pytest_asyncio

10 import torch.cuda

11

12 +from tests.utils import RemoteOpenAIServer

13 from vllm.engine.arg_utils import EngineArgs

14 from vllm.model_executor.model_loader.tensorizer import (

15 TensorizerConfig,

@@ -17,8 +18,6 @@ from vllm.model_executor.model_loader.tensorizer import (

18 )

19 from vllm.platforms import current_platform

20

20 -from ...utils import RemoteOpenAIServer

21 -

21 MODEL_NAME = "unsloth/llama-3.2-1b-Instruct"

22 LORA_PATH = "davzoku/finqa_adapter_1b"

23

tests/entrypoints/openai/completion/test_token_in_token_out.py renamed

+1 -2

@@ -6,11 +6,10 @@ import tempfile

6

7 import pytest

8

9 +from tests.utils import RemoteOpenAIServer

10 from vllm.model_executor.model_loader.weight_utils import download_weights_from_hf

11 from vllm.tokenizers import get_tokenizer

12

12 -from ...utils import RemoteOpenAIServer

13 -

13 MODEL_NAME = "Qwen/Qwen3-0.6B"

14 MODEL_PATH = os.path.join(tempfile.gettempdir(), "qwen3_06b")

15

[Refactor] Relocate completion and chat completion tests (#37125)