[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
@@ -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