A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 22:45 UTC
[BugFix]: Fix local mypy issues (#34739)
Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Martin Hickey committed
Feb 23, 2026 at 08:40 UTC
e97c46a92ded70813806fd6f33cf8bd9fcf005bc
6 files changed
+12
-6
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
+4
@@ -935,6 +935,10 @@ class NixlConnectorWorker:
935
]
936
937
if rsv_cores_for_kv:
938
+ if not hasattr(os, "sched_setaffinity"):
939
+ raise NotImplementedError(
940
+ "os.sched_setaffinity is not available on this platform"
941
+ )
942
os.sched_setaffinity(0, rsv_cores_for_kv)
943
944
# support for oot platform which can't register nixl memory
vllm/entrypoints/openai/chat_completion/protocol.py
+1
-1
@@ -5,7 +5,6 @@
5
# https://github.com/lm-sys/FastChat/blob/168ccc29d3f7edc50823016105c024fe2282732a/fastchat/protocol/openai_api_protocol.py
6
import json
7
import time
8
-from dataclasses import replace
8
from typing import Annotated, Any, ClassVar, Literal
9
10
import torch
@@ -16,6 +15,7 @@ from openai.types.chat.chat_completion_message import Annotation as OpenAIAnnota
15
from pydantic import Field, model_validator
16
17
from vllm.config import ModelConfig
18
+from vllm.config.utils import replace
19
from vllm.entrypoints.chat_utils import (
20
ChatCompletionMessageParam,
21
ChatTemplateContentFormatOption,
vllm/entrypoints/openai/completion/protocol.py
+1
-1
@@ -5,13 +5,13 @@
5
# https://github.com/lm-sys/FastChat/blob/168ccc29d3f7edc50823016105c024fe2282732a/fastchat/protocol/openai_api_protocol.py
6
import json
7
import time
8
-from dataclasses import replace
8
from typing import Annotated, Any, Literal
9
10
import torch
11
from pydantic import Field, model_validator
12
13
from vllm.config import ModelConfig
14
+from vllm.config.utils import replace
15
from vllm.entrypoints.openai.engine.protocol import (
16
AnyResponseFormat,
17
LegacyStructuralTagResponseFormat,
vllm/entrypoints/openai/responses/protocol.py
+3
-1
@@ -337,7 +337,9 @@ class ResponsesRequest(OpenAIBaseModel):
337
and response_format.schema_ is not None
338
):
339
structured_outputs = StructuredOutputsParams(
340
- json=response_format.schema_
340
+ json=response_format.schema_ # type: ignore[call-arg]
341
+ # --follow-imports skip hides the class definition but also hides
342
+ # multiple third party conflicts, so best of both evils
343
)
344
345
stop = self.stop if self.stop else []
vllm/entrypoints/openai/responses/serving.py
+1
-1
@@ -8,7 +8,6 @@ from collections import deque
8
from collections.abc import AsyncGenerator, AsyncIterator, Callable, Sequence
9
from contextlib import AsyncExitStack
10
from copy import copy
11
-from dataclasses import replace
11
from http import HTTPStatus
12
from typing import Final
13
@@ -40,6 +39,7 @@ from openai_harmony import Message as OpenAIHarmonyMessage
39
from pydantic import TypeAdapter
40
41
from vllm import envs
42
+from vllm.config.utils import replace
43
from vllm.engine.protocol import EngineClient
44
from vllm.entrypoints.chat_utils import (
45
ChatCompletionMessageParam,
vllm/sampling_params.py
+2
-2
@@ -3,7 +3,7 @@
3
"""Sampling parameters for text generation."""
4
5
import copy
6
-import json
6
+import json as json_mod
7
from dataclasses import field
8
from enum import Enum, IntEnum
9
from functools import cached_property
@@ -791,7 +791,7 @@ class SamplingParams(
791
skip_guidance = False
792
if so_params.json:
793
if isinstance(so_params.json, str):
794
- schema = json.loads(so_params.json)
794
+ schema = json_mod.loads(so_params.json)
795
else:
796
schema = so_params.json
797
skip_guidance = has_guidance_unsupported_json_features(schema)