[MyPy][3/N] Fix MyPy errors in test groups (part 3) (#55939)

Signed-off-by: Taimys <ffy2281693949@gmail.com>
Signed-off-by: Luyi Xiao <81631911+luyixiao95@users.noreply.github.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Luyi Xiao <81631911+luyixiao95@users.noreply.github.com>
Co-authored-by: Taimys <ffy2281693949@gmail.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Taimys
2026-09-30 11:06:07 +00:00
committed by GitHub
co-authored by Codex Luyi Xiao Taimys Harry Mellor Claude Opus 5.5
parent c4fa6f36a7
commit 5225f1bacb
118 changed files with 948 additions and 605 deletions
@@ -355,9 +355,11 @@ class TestToolResultContent:
if m["role"] == "user" and isinstance(m.get("content"), list)
]
assert len(follow_up) == 1
assert follow_up[0]["content"][0]["image_url"]["url"] == (
"data:image/jpeg;base64,QUFB"
)
content = follow_up[0]["content"]
assert isinstance(content, list)
image = content[0]
assert image["type"] == "image_url"
assert image["image_url"]["url"] == "data:image/jpeg;base64,QUFB"
def test_tool_result_with_multiple_images(self):
request = self._make_tool_result_request(
@@ -387,7 +389,12 @@ class TestToolResultContent:
if m["role"] == "user" and isinstance(m.get("content"), list)
]
assert len(follow_up) == 1
urls = [p["image_url"]["url"] for p in follow_up[0]["content"]]
content = follow_up[0]["content"]
assert isinstance(content, list)
urls = []
for part in content:
assert part["type"] == "image_url"
urls.append(part["image_url"]["url"])
assert urls == [
"data:image/png;base64,IMG1",
"https://example.com/img2.jpg",
@@ -584,7 +591,8 @@ class TestThinkingBlockConversion:
asst = asst_msgs[0]
assert asst.get("reasoning") == "I need to call the calculator."
tool_calls = list(asst.get("tool_calls", []))
tool_calls = asst.get("tool_calls")
assert tool_calls is not None
assert len(tool_calls) == 1
assert tool_calls[0]["function"]["name"] == "calculator"
# No text content alongside reasoning + tool_use.
@@ -1439,7 +1447,7 @@ class TestDetectMergeInlineSystem:
model_config=None,
)
assert (
AnthropicServingMessages._should_merge_inline_system(online_renderer)
AnthropicServingMessages._should_merge_inline_system(online_renderer) # type: ignore[arg-type]
is expected
)
assert mock_logger.warning_once.called is expected
@@ -7,22 +7,26 @@ from unittest.mock import Mock
import pytest
from vllm import LLM, SamplingParams
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
from vllm.inputs import TextPrompt
from vllm.outputs import RequestOutput
def _make_mock_llm() -> LLM:
llm = object.__new__(LLM)
llm.model_config = SimpleNamespace(
llm.model_config = SimpleNamespace( # type: ignore[assignment]
runner_type="generate", enable_prompt_embeds=False
)
return llm
def test_generate_forwards_mm_processor_kwargs() -> None:
def test_generate_forwards_mm_processor_kwargs(monkeypatch: pytest.MonkeyPatch) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"num_crops": 4}
sampling_params = SamplingParams(max_tokens=1)
llm._run_completion = Mock(return_value=["ok"])
_run_completion = Mock(return_value=["ok"])
monkeypatch.setattr(llm, "_run_completion", _run_completion)
outputs = llm.generate(
"prompt",
@@ -31,17 +35,18 @@ def test_generate_forwards_mm_processor_kwargs() -> None:
)
assert outputs == ["ok"]
assert llm._run_completion.call_args.kwargs["mm_processor_kwargs"] == (
assert _run_completion.call_args.kwargs["mm_processor_kwargs"] == (
mm_processor_kwargs
)
def test_enqueue_forwards_mm_processor_kwargs() -> None:
def test_enqueue_forwards_mm_processor_kwargs(monkeypatch: pytest.MonkeyPatch) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"do_resize": False}
sampling_params = SamplingParams(max_tokens=1)
llm._add_completion_requests = Mock(return_value=["req-0"])
_add_completion_requests = Mock(return_value=["req-0"])
monkeypatch.setattr(llm, "_add_completion_requests", _add_completion_requests)
request_ids = llm.enqueue(
"prompt",
@@ -51,18 +56,19 @@ def test_enqueue_forwards_mm_processor_kwargs() -> None:
)
assert request_ids == ["req-0"]
assert llm._add_completion_requests.call_args.kwargs["mm_processor_kwargs"] == (
assert _add_completion_requests.call_args.kwargs["mm_processor_kwargs"] == (
mm_processor_kwargs
)
def test_chat_forwards_mm_processor_kwargs() -> None:
def test_chat_forwards_mm_processor_kwargs(monkeypatch: pytest.MonkeyPatch) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"do_pan_and_scan": True}
sampling_params = SamplingParams(max_tokens=1)
messages = [{"role": "user", "content": "hello"}]
messages: list[ChatCompletionMessageParam] = [{"role": "user", "content": "hello"}]
llm._run_chat = Mock(return_value=["ok"])
_run_chat = Mock(return_value=["ok"])
monkeypatch.setattr(llm, "_run_chat", _run_chat)
outputs = llm.chat(
messages,
@@ -71,18 +77,19 @@ def test_chat_forwards_mm_processor_kwargs() -> None:
)
assert outputs == ["ok"]
assert llm._run_chat.call_args.kwargs["mm_processor_kwargs"] == (
mm_processor_kwargs
)
assert _run_chat.call_args.kwargs["mm_processor_kwargs"] == mm_processor_kwargs
def test_enqueue_chat_forwards_mm_processor_kwargs() -> None:
def test_enqueue_chat_forwards_mm_processor_kwargs(
monkeypatch: pytest.MonkeyPatch,
) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"do_pan_and_scan": True}
sampling_params = SamplingParams(max_tokens=1)
messages = [{"role": "user", "content": "hello"}]
messages: list[ChatCompletionMessageParam] = [{"role": "user", "content": "hello"}]
llm._add_chat_requests = Mock(return_value=["req-0"])
_add_chat_requests = Mock(return_value=["req-0"])
monkeypatch.setattr(llm, "_add_chat_requests", _add_chat_requests)
request_ids = llm.enqueue_chat(
messages,
@@ -92,67 +99,79 @@ def test_enqueue_chat_forwards_mm_processor_kwargs() -> None:
)
assert request_ids == ["req-0"]
assert llm._add_chat_requests.call_args.kwargs["mm_processor_kwargs"] == (
assert _add_chat_requests.call_args.kwargs["mm_processor_kwargs"] == (
mm_processor_kwargs
)
def test_run_chat_forwards_mm_processor_kwargs() -> None:
def test_run_chat_forwards_mm_processor_kwargs(monkeypatch: pytest.MonkeyPatch) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"num_crops": 8}
sampling_params = SamplingParams(max_tokens=1)
messages = [{"role": "user", "content": "hello"}]
messages: list[ChatCompletionMessageParam] = [{"role": "user", "content": "hello"}]
sentinel_output = ["done"]
llm._add_chat_requests = Mock()
llm._run_engine = Mock(return_value=sentinel_output)
_add_chat_requests = Mock()
monkeypatch.setattr(llm, "_add_chat_requests", _add_chat_requests)
_run_engine = Mock(return_value=sentinel_output)
monkeypatch.setattr(llm, "_run_engine", _run_engine)
outputs = llm._run_chat(
messages=messages,
params=sampling_params,
output_type=object,
output_type=RequestOutput,
use_tqdm=False,
mm_processor_kwargs=mm_processor_kwargs,
)
assert outputs == sentinel_output
assert llm._add_chat_requests.call_args.kwargs["mm_processor_kwargs"] == (
assert _add_chat_requests.call_args.kwargs["mm_processor_kwargs"] == (
mm_processor_kwargs
)
def test_run_completion_forwards_mm_processor_kwargs() -> None:
def test_run_completion_forwards_mm_processor_kwargs(
monkeypatch: pytest.MonkeyPatch,
) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"min_pixels": 4 * 28 * 28}
sampling_params = SamplingParams(max_tokens=1)
sentinel_output = ["done"]
llm._add_completion_requests = Mock()
llm._run_engine = Mock(return_value=sentinel_output)
_add_completion_requests = Mock()
monkeypatch.setattr(llm, "_add_completion_requests", _add_completion_requests)
_run_engine = Mock(return_value=sentinel_output)
monkeypatch.setattr(llm, "_run_engine", _run_engine)
outputs = llm._run_completion(
prompts=["prompt"],
params=sampling_params,
output_type=object,
output_type=RequestOutput,
use_tqdm=False,
mm_processor_kwargs=mm_processor_kwargs,
)
assert outputs == sentinel_output
assert llm._add_completion_requests.call_args.kwargs["mm_processor_kwargs"] == (
assert _add_completion_requests.call_args.kwargs["mm_processor_kwargs"] == (
mm_processor_kwargs
)
def test_add_completion_requests_forwards_mm_processor_kwargs() -> None:
def test_add_completion_requests_forwards_mm_processor_kwargs(
monkeypatch: pytest.MonkeyPatch,
) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"max_dynamic_patch": 4}
sampling_params = SamplingParams(max_tokens=1)
llm._params_to_seq = Mock(return_value=[sampling_params])
llm._lora_request_to_seq = Mock(return_value=[None])
llm._priority_to_seq = Mock(return_value=[0])
llm._preprocess_cmpl_one = Mock(return_value={"prompt_token_ids": [1]})
_params_to_seq = Mock(return_value=[sampling_params])
monkeypatch.setattr(llm, "_params_to_seq", _params_to_seq)
_lora_request_to_seq = Mock(return_value=[None])
monkeypatch.setattr(llm, "_lora_request_to_seq", _lora_request_to_seq)
_priority_to_seq = Mock(return_value=[0])
monkeypatch.setattr(llm, "_priority_to_seq", _priority_to_seq)
_preprocess_cmpl_one = Mock(return_value={"prompt_token_ids": [1]})
monkeypatch.setattr(llm, "_preprocess_cmpl_one", _preprocess_cmpl_one)
captured_prompts = []
@@ -160,7 +179,8 @@ def test_add_completion_requests_forwards_mm_processor_kwargs() -> None:
captured_prompts.extend(prompts)
return ["req-0"]
llm._render_and_add_requests = Mock(side_effect=fake_render_and_add_requests)
_render_and_add_requests = Mock(side_effect=fake_render_and_add_requests)
monkeypatch.setattr(llm, "_render_and_add_requests", _render_and_add_requests)
request_ids = llm._add_completion_requests(
prompts=["prompt"],
@@ -170,7 +190,7 @@ def test_add_completion_requests_forwards_mm_processor_kwargs() -> None:
)
assert request_ids == ["req-0"]
llm._preprocess_cmpl_one.assert_called_once_with(
_preprocess_cmpl_one.assert_called_once_with(
"prompt",
None,
mm_processor_kwargs=mm_processor_kwargs,
@@ -183,7 +203,7 @@ def test_preprocess_cmpl_applies_mm_processor_kwargs_to_renderer(
) -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"num_crops": 8}
prompt = {"prompt": "<image>", "multi_modal_data": {"image": object()}}
prompt: TextPrompt = {"prompt": "<image>", "multi_modal_data": {"image": object()}}
renderer = Mock()
renderer.default_cmpl_tok_params = Mock()
@@ -213,7 +233,7 @@ def test_preprocess_cmpl_keeps_prompt_mm_processor_kwargs_when_no_override(
monkeypatch: pytest.MonkeyPatch,
) -> None:
llm = _make_mock_llm()
prompt = {
prompt: TextPrompt = {
"prompt": "<image>",
"multi_modal_data": {"image": object()},
"mm_processor_kwargs": {"num_crops": 2},
@@ -243,7 +263,9 @@ def test_preprocess_cmpl_keeps_prompt_mm_processor_kwargs_when_no_override(
def test_preprocess_chat_applies_mm_processor_kwargs_to_renderer() -> None:
llm = _make_mock_llm()
mm_processor_kwargs = {"num_crops": 8}
messages = [[{"role": "user", "content": "Describe this image."}]]
messages: list[list[ChatCompletionMessageParam]] = [
[{"role": "user", "content": "Describe this image."}]
]
renderer = Mock()
renderer.tokenizer = object()
@@ -269,7 +291,9 @@ def test_preprocess_chat_applies_mm_processor_kwargs_to_renderer() -> None:
def test_preprocess_chat_omits_mm_processor_kwargs_when_no_override() -> None:
llm = _make_mock_llm()
messages = [[{"role": "user", "content": "Describe this image."}]]
messages: list[list[ChatCompletionMessageParam]] = [
[{"role": "user", "content": "Describe this image."}]
]
renderer = Mock()
renderer.tokenizer = object()
@@ -291,7 +315,9 @@ def test_preprocess_chat_omits_mm_processor_kwargs_when_no_override() -> None:
def test_preprocess_chat_defaults_add_special_tokens_to_false() -> None:
# Matches `ChatCompletionRequest.add_special_tokens` on the server.
llm = _make_mock_llm()
messages = [[{"role": "user", "content": "hi"}]]
messages: list[list[ChatCompletionMessageParam]] = [
[{"role": "user", "content": "hi"}]
]
renderer = Mock()
renderer.tokenizer = object()
@@ -309,7 +335,9 @@ def test_preprocess_chat_defaults_add_special_tokens_to_false() -> None:
def test_preprocess_chat_tokenization_kwargs_override_add_special_tokens() -> None:
llm = _make_mock_llm()
messages = [[{"role": "user", "content": "hi"}]]
messages: list[list[ChatCompletionMessageParam]] = [
[{"role": "user", "content": "hi"}]
]
renderer = Mock()
renderer.tokenizer = object()
@@ -47,18 +47,20 @@ async def client(server):
@pytest.fixture(scope="session")
def base64_encoded_audio() -> dict[str, str]:
return {
audio_url: encode_audio_base64(*fetch_audio(audio_url))
for audio_url in TEST_AUDIO_URLS
}
encoded = {}
for audio_url in TEST_AUDIO_URLS:
audio, sample_rate = fetch_audio(audio_url)
encoded[audio_url] = encode_audio_base64(audio, int(sample_rate))
return encoded
@pytest.fixture(scope="session")
def url_encoded_audio() -> dict[str, str]:
return {
audio_url: encode_audio_url(*fetch_audio(audio_url))
for audio_url in TEST_AUDIO_URLS
}
encoded = {}
for audio_url in TEST_AUDIO_URLS:
audio, sample_rate = fetch_audio(audio_url)
encoded[audio_url] = encode_audio_url(audio, int(sample_rate))
return encoded
def dummy_messages_from_audio_url(
@@ -7,6 +7,7 @@ import torch
from tests.models.utils import softmax
from vllm import LLM, ClassificationRequestOutput, PoolingParams
from vllm.inputs import PromptType
from vllm.tasks import PoolingTask
MODEL_NAME = "jason9693/Qwen2.5-1.5B-apeach"
@@ -53,7 +54,8 @@ def test_token_ids_prompts(llm: LLM):
@pytest.mark.skip_global_cleanup
def test_list_prompts(llm: LLM):
outputs = llm.classify([prompt, prompt_token_ids], use_tqdm=False)
prompts: list[PromptType] = [prompt, prompt_token_ids]
outputs = llm.classify(prompts, use_tqdm=False)
assert len(outputs) == 2
for i in range(len(outputs)):
assert isinstance(outputs[i], ClassificationRequestOutput)
@@ -21,6 +21,7 @@ from vllm.entrypoints.pooling.typing import (
PoolingEngineInput,
PoolingServeContext,
)
from vllm.inputs import tokens_input
from vllm.outputs import PoolingOutput, PoolingRequestOutput
@@ -28,7 +29,7 @@ class TestEmbeddingRequestParsing:
"""Unit tests for OpenAI embedding request parsing."""
def test_input_messages_parses_as_chat_request(self):
request = TypeAdapter(EmbeddingRequest).validate_python(
request: EmbeddingRequest = TypeAdapter(EmbeddingRequest).validate_python(
{
"model": "test",
"input": [{"role": "user", "content": "hello"}],
@@ -42,7 +43,7 @@ class TestEmbeddingRequestParsing:
assert request.chat_template_kwargs == {"instruction": "Represent the query: "}
def test_batched_input_messages_parses_as_batch_chat_input_request(self):
request = TypeAdapter(EmbeddingRequest).validate_python(
request: EmbeddingRequest = TypeAdapter(EmbeddingRequest).validate_python(
{
"model": "test",
"input": [
@@ -65,7 +66,7 @@ class TestEmbeddingRequestParsing:
assert request.chat_template_kwargs == {"instruction": "Represent the query: "}
def test_token_ids_still_parse_as_completion_request(self):
request = TypeAdapter(EmbeddingRequest).validate_python(
request: EmbeddingRequest = TypeAdapter(EmbeddingRequest).validate_python(
{
"model": "test",
"input": [[1, 2, 3], [4, 5]],
@@ -76,7 +77,7 @@ class TestEmbeddingRequestParsing:
assert request.input == [[1, 2, 3], [4, 5]]
def test_messages_still_parses_as_chat_request(self):
request = TypeAdapter(EmbeddingRequest).validate_python(
request: EmbeddingRequest = TypeAdapter(EmbeddingRequest).validate_python(
{
"model": "test",
"messages": [{"role": "user", "content": "hello"}],
@@ -89,7 +90,7 @@ class TestEmbeddingRequestParsing:
assert request.chat_template_kwargs == {"instruction": "Represent the query: "}
def test_batched_messages_parses_as_batch_chat_request(self):
request = TypeAdapter(EmbeddingRequest).validate_python(
request: EmbeddingRequest = TypeAdapter(EmbeddingRequest).validate_python(
{
"model": "test",
"messages": [
@@ -410,13 +411,13 @@ class TestChunkedEmbeddingProcessing:
@classmethod
def _make_handler(cls):
handler = object.__new__(EmbedIOProcessor)
handler.model_config = cls._FakeModelConfig()
handler.model_config = cls._FakeModelConfig() # type: ignore[assignment]
handler.enable_chunked_processing = True
return handler
@staticmethod
def _make_context() -> PoolingServeContext[EmbeddingCompletionRequest]:
request = TypeAdapter(EmbeddingRequest).validate_python(
request: EmbeddingRequest = TypeAdapter(EmbeddingRequest).validate_python(
{
"model": "test",
"input": [[0, 1, 2, 3, 4], [10, 11]],
@@ -431,13 +432,13 @@ class TestChunkedEmbeddingProcessing:
request_id="embd-client-prompt-999-chunk-888",
engine_inputs=[
PoolingEngineInput(
prompts={"prompt_token_ids": [0, 1, 2, 3, 4]},
prompts=tokens_input(prompt_token_ids=[0, 1, 2, 3, 4]),
params=pooling_params,
lora_requests=None,
priorities=0,
),
PoolingEngineInput(
prompts={"prompt_token_ids": [10, 11]},
prompts=tokens_input(prompt_token_ids=[10, 11]),
params=pooling_params,
lora_requests=None,
priorities=0,
@@ -7,6 +7,7 @@ import torch
import torch.nn.functional as F
from vllm import LLM, EmbeddingRequestOutput, PoolingParams
from vllm.inputs import PromptType
from vllm.tasks import PoolingTask
MODEL_NAME = "intfloat/multilingual-e5-small"
@@ -54,7 +55,8 @@ def test_token_ids_prompts(llm: LLM):
@pytest.mark.skip_global_cleanup
def test_list_prompts(llm: LLM):
outputs = llm.embed([prompt, prompt_token_ids], use_tqdm=False)
prompts: list[PromptType] = [prompt, prompt_token_ids]
outputs = llm.embed(prompts, use_tqdm=False)
assert len(outputs) == 2
for i in range(len(outputs)):
assert isinstance(outputs[i], EmbeddingRequestOutput)
+13 -9
View File
@@ -12,9 +12,13 @@ import requests
import torch
import torch.nn.functional as F
from tests.models.language.pooling.embed_utils import run_embedding_correctness_test
from tests.models.language.pooling.embed_utils import (
float_embeddings,
run_embedding_correctness_test,
)
from tests.models.utils import check_embeddings_close
from tests.utils import RemoteOpenAIServer
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
from vllm.entrypoints.pooling.embed.protocol import EmbeddingResponse
from vllm.entrypoints.pooling.pooling.protocol import PoolingResponse
from vllm.entrypoints.pooling.utils import (
@@ -145,7 +149,7 @@ async def test_completion_request(
assert embeddings.usage.prompt_tokens == len(input_tokens)
assert embeddings.usage.total_tokens == len(input_tokens)
vllm_outputs = [d.embedding for d in embeddings.data]
vllm_outputs = float_embeddings(embeddings)
run_embedding_correctness_test(hf_model, [input_text], vllm_outputs)
# test input: list[int]
@@ -165,7 +169,7 @@ async def test_completion_request(
assert embeddings.usage.prompt_tokens == len(input_tokens)
assert embeddings.usage.total_tokens == len(input_tokens)
vllm_outputs = [d.embedding for d in embeddings.data]
vllm_outputs = float_embeddings(embeddings)
run_embedding_correctness_test(hf_model, [input_text], vllm_outputs)
@@ -194,7 +198,7 @@ async def test_completion_request_batched(
assert embeddings.usage.prompt_tokens == len(input_tokens) * N
assert embeddings.usage.total_tokens == len(input_tokens) * N
vllm_outputs = [d.embedding for d in embeddings.data]
vllm_outputs = float_embeddings(embeddings)
run_embedding_correctness_test(hf_model, input_texts, vllm_outputs)
# test list[list[int]]
@@ -214,7 +218,7 @@ async def test_completion_request_batched(
assert embeddings.usage.prompt_tokens == len(input_tokens) * N
assert embeddings.usage.total_tokens == len(input_tokens) * N
vllm_outputs = [d.embedding for d in embeddings.data]
vllm_outputs = float_embeddings(embeddings)
run_embedding_correctness_test(hf_model, input_texts, vllm_outputs)
@@ -331,7 +335,7 @@ async def test_padding(client: openai.AsyncOpenAI, model_name: str):
async def test_chat_request(
server: RemoteOpenAIServer, client: openai.AsyncOpenAI, model_name: str
):
messages = [
messages: list[ChatCompletionMessageParam] = [
{
"role": "user",
"content": "The cat sat on the mat.",
@@ -382,8 +386,8 @@ async def test_chat_request(
assert chat_embeddings.created <= completion_embeddings.created
# Use tolerance-based comparison for embeddings
check_embeddings_close(
embeddings_0_lst=[d.embedding for d in chat_embeddings.data],
embeddings_1_lst=[d.embedding for d in completion_embeddings.data],
embeddings_0_lst=float_embeddings(chat_embeddings),
embeddings_1_lst=float_embeddings(completion_embeddings),
name_0="chat",
name_1="completion",
)
@@ -488,7 +492,7 @@ async def test_invocations_completion_request(
@pytest.mark.asyncio
async def test_invocations_chat_request(server: RemoteOpenAIServer):
messages = [
messages: list[ChatCompletionMessageParam] = [
{
"role": "user",
"content": "The cat sat on the mat.",
@@ -6,7 +6,10 @@ import openai
import pytest
from tests.conftest import HfRunner
from tests.models.language.pooling.embed_utils import run_embedding_correctness_test
from tests.models.language.pooling.embed_utils import (
float_embeddings,
run_embedding_correctness_test,
)
from tests.models.utils import EmbedModelInfo
from tests.utils import ROCM_EXTRA_ARGS, RemoteOpenAIServer
from vllm.entrypoints.pooling.embed.protocol import EmbeddingResponse
@@ -95,7 +98,7 @@ async def test_matryoshka(
if dimensions is not None:
assert len(embeddings.data[0].embedding) == dimensions
vllm_outputs = [d.embedding for d in embeddings.data]
vllm_outputs = float_embeddings(embeddings)
run_embedding_correctness_test(hf_model, prompts, vllm_outputs, dimensions)
if model_info.is_matryoshka:
@@ -274,6 +274,7 @@ async def test_long_text_embedding_1500_chars(
# Verify embedding vector validity
embedding_vector = embeddings.data[0].embedding
assert isinstance(embedding_vector, list)
assert all(isinstance(x, float) for x in embedding_vector), (
"Embedding vector should contain floats"
)
@@ -322,6 +323,7 @@ async def test_long_text_embedding_2500_chars(
# Verify embedding vector validity
embedding_vector = embeddings.data[0].embedding
assert isinstance(embedding_vector, list)
assert all(isinstance(x, float) for x in embedding_vector), (
"Embedding vector should contain floats"
)
@@ -409,6 +411,7 @@ async def test_chunked_vs_normal_consistency(
# 验证embedding向量的有效性
embedding_vector = embeddings.data[0].embedding
assert isinstance(embedding_vector, list)
assert all(isinstance(x, float) for x in embedding_vector)
assert not all(x == 0 for x in embedding_vector)
@@ -438,6 +441,7 @@ async def test_chunked_processing_response_format(
# Verify embedding vector properties
embedding_vector = embeddings.data[0].embedding
assert isinstance(embedding_vector, list)
import math
vector_norm = math.sqrt(sum(x * x for x in embedding_vector))
@@ -11,6 +11,7 @@ import torch
from tests.models.utils import check_embeddings_close
from tests.utils import RemoteOpenAIServer
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
from vllm.entrypoints.pooling.pooling.protocol import PoolingResponse
from vllm.entrypoints.pooling.utils import (
MetadataItem,
@@ -137,7 +138,7 @@ def test_completion_request_batched(server: RemoteOpenAIServer, model_name: str)
@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
async def test_chat_request(server: RemoteOpenAIServer, model_name: str):
messages = [
messages: list[ChatCompletionMessageParam] = [
{
"role": "user",
"content": "The cat sat on the mat.",
@@ -524,7 +525,7 @@ async def test_invocations_chat_request(server: RemoteOpenAIServer):
@pytest.mark.asyncio
async def test_invocations_conversation_chat_request(server: RemoteOpenAIServer):
messages = [
messages: list[ChatCompletionMessageParam] = [
{
"role": "user",
"content": "The cat sat on the mat.",
@@ -47,7 +47,7 @@ def llm_reranker_processor() -> CrossEncoderIOProcessor:
)
processor = CrossEncoderIOProcessor.__new__(CrossEncoderIOProcessor)
processor.model_config = SimpleNamespace(enable_prompt_embeds=False)
processor.model_config = SimpleNamespace(enable_prompt_embeds=False) # type: ignore[assignment]
processor.tokenizer = TokenizersBackend(
tokenizer_object=backend,
unk_token="[UNK]",
@@ -95,12 +95,16 @@ def test_token_type_ids_stay_aligned_with_a_truncated_padded_prompt():
truncation_side="left",
)
prompt = tok_params.apply_post_tokenization(tokenizer, prompt)
processed_prompt = tok_params.apply_post_tokenization(
tokenizer, # type: ignore[arg-type]
prompt,
)
assert "prompt_token_ids" in processed_prompt
token_type_ids = _apply_post_tokenization_to_token_type_ids(
tokenizer, tok_params, token_type_ids
)
prompt_token_ids = prompt["prompt_token_ids"]
prompt_token_ids = processed_prompt["prompt_token_ids"]
assert len(token_type_ids) == len(prompt_token_ids)
# Keeping the last 40 tokens drops the first 10 query tokens, so 10 query
@@ -30,12 +30,18 @@ def test_online_forwards_truncate_prompt_tokens_to_proxy(monkeypatch):
silently disables truncate_prompt_tokens for Jina rerank/score.
"""
proc = JinaRankingIOProcessor.__new__(JinaRankingIOProcessor)
proc.valid_inputs_online = MagicMock(
return_value=ScoringData(data_1=["query"], data_2=["doc"])
monkeypatch.setattr(
proc,
"valid_inputs_online",
MagicMock(return_value=ScoringData(data_1=["query"], data_2=["doc"])),
)
monkeypatch.setattr(proc, "_get_token_limits", MagicMock(return_value=(0, 0)))
monkeypatch.setattr(
proc, "ensure_str", MagicMock(side_effect=lambda data: list(data))
)
monkeypatch.setattr(
proc, "format_docs_prompts_func", MagicMock(return_value="formatted prompt")
)
proc._get_token_limits = MagicMock(return_value=(0, 0))
proc.ensure_str = MagicMock(side_effect=lambda data: list(data))
proc.format_docs_prompts_func = MagicMock(return_value="formatted prompt")
captured: dict[str, object] = {}
@@ -71,8 +77,10 @@ def offline_processor_and_context():
backend.pre_tokenizer = pre_tokenizers.WhitespaceSplit()
proc = JinaRankingIOProcessor.__new__(JinaRankingIOProcessor)
proc.tokenizer = TokenizersBackend(tokenizer_object=backend)
proc.model_config = SimpleNamespace(max_model_len=1024, is_encoder_decoder=False)
proc.renderer = SimpleNamespace(
proc.model_config = SimpleNamespace( # type: ignore[assignment]
max_model_len=1024, is_encoder_decoder=False
)
proc.renderer = SimpleNamespace( # type: ignore[assignment]
default_cmpl_tok_params=TokenizeParams(max_total_tokens=1024)
)
ctx = OfflineScoringInputsContext(
@@ -6,7 +6,7 @@ from unittest.mock import AsyncMock, Mock
import pytest
from vllm.entrypoints.pooling.scoring.serving import ServingScores
from vllm.entrypoints.pooling.typing import PoolingServeContext
from vllm.entrypoints.pooling.typing import PoolingEngineInput, PoolingServeContext
from vllm.inputs import tokens_input
from vllm.pooling_params import PoolingParams
@@ -21,8 +21,18 @@ def _make_context() -> PoolingServeContext:
priorities=None,
prompt_extras=None,
engine_inputs=[
tokens_input(prompt_token_ids=[1]),
tokens_input(prompt_token_ids=[2]),
PoolingEngineInput(
prompts=tokens_input(prompt_token_ids=[1]),
params=PoolingParams(task="token_embed"),
lora_requests=None,
priorities=0,
),
PoolingEngineInput(
prompts=tokens_input(prompt_token_ids=[2]),
params=PoolingParams(task="token_embed"),
lora_requests=None,
priorities=0,
),
],
n_queries=1,
)
@@ -39,10 +49,13 @@ def _query_key(context: PoolingServeContext) -> str:
@pytest.mark.asyncio
async def test_colliding_request_ids_use_distinct_query_cache_keys():
async def test_colliding_request_ids_use_distinct_query_cache_keys(
monkeypatch: pytest.MonkeyPatch,
):
serving = object.__new__(ServingScores)
serving._prepare_generators = AsyncMock()
serving._collect_batch = AsyncMock()
prepare_generators = AsyncMock()
monkeypatch.setattr(serving, "_prepare_generators", prepare_generators)
monkeypatch.setattr(serving, "_collect_batch", AsyncMock())
first = _make_context()
second = _make_context()
@@ -50,9 +63,7 @@ async def test_colliding_request_ids_use_distinct_query_cache_keys():
await serving._flash_late_interaction_encode_queries(context)
await serving._flash_late_interaction_encode_docs(context)
prepared_contexts = [
call.args[0] for call in serving._prepare_generators.await_args_list
]
prepared_contexts = [call.args[0] for call in prepare_generators.await_args_list]
first_query_key = _query_key(prepared_contexts[0])
first_doc_key = _query_key(prepared_contexts[1])
second_query_key = _query_key(prepared_contexts[2])
@@ -53,8 +53,8 @@ def test_rejects_mismatched_lora_requests(processor: PoolingIOProcessor):
def test_rejects_conflicting_pooling_task(processor: PoolingIOProcessor):
processor.model_config = SimpleNamespace(is_encoder_decoder=False)
processor.renderer = SimpleNamespace(
processor.model_config = SimpleNamespace(is_encoder_decoder=False) # type: ignore[assignment]
processor.renderer = SimpleNamespace( # type: ignore[assignment] # Tokenization defaults only.
default_cmpl_tok_params=TokenizeParams(max_total_tokens=None)
)
ctx = OfflineEncodeInputsContext(
@@ -45,6 +45,7 @@ async def test_pooling_token_classify(server: RemoteOpenAIServer, model_name: st
poolings = PoolingResponse.model_validate(response.json())
assert len(poolings.data) == 1
assert len(poolings.data[0].data) == 8
assert isinstance(poolings.data[0].data[0], list)
assert len(poolings.data[0].data[0]) == 2
@@ -71,6 +71,7 @@ async def test_pooling_token_embed(server: RemoteOpenAIServer, model_name: str):
assert poolings.id == f"pooling-{request_id}"
assert len(poolings.data) == 1
assert len(poolings.data[0].data) == len(input_tokens)
assert isinstance(poolings.data[0].data[0], list)
assert len(poolings.data[0].data[0]) == 384