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 16, 00:45 UTC
Fix the pydantic logging validator (#12420)
Signed-off-by: Max de Bayser <mbayser@br.ibm.com>
Maximilien de Bayser committed
Jan 29, 2025 at 04:53 UTC
ef001d98ef36166ebacb48eab2e32eb738407b53
1 file changed
+22
-7
vllm/entrypoints/openai/protocol.py
+22
-7
@@ -6,7 +6,8 @@ from argparse import Namespace
6
from typing import Any, ClassVar, Dict, List, Literal, Optional, Set, Union
7
8
import torch
9
-from pydantic import BaseModel, ConfigDict, Field, model_validator
9
+from pydantic import (BaseModel, ConfigDict, Field, TypeAdapter,
10
+ ValidationInfo, field_validator, model_validator)
11
from typing_extensions import Annotated
12
13
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
@@ -45,14 +46,14 @@ class OpenAIBaseModel(BaseModel):
46
# Cache class field names
47
field_names: ClassVar[Optional[Set[str]]] = None
48
48
- @model_validator(mode="before")
49
+ @model_validator(mode="wrap")
50
@classmethod
50
- def __log_extra_fields__(cls, data):
51
-
51
+ def __log_extra_fields__(cls, data, handler):
52
+ result = handler(data)
53
+ if not isinstance(data, dict):
54
+ return result
55
field_names = cls.field_names
56
if field_names is None:
54
- if not isinstance(data, dict):
55
- return data
57
# Get all class field names and their potential aliases
58
field_names = set()
59
for field_name, field in cls.model_fields.items():
@@ -67,7 +68,7 @@ class OpenAIBaseModel(BaseModel):
68
"The following fields were present in the request "
69
"but ignored: %s",
70
data.keys() - field_names)
70
- return data
71
+ return result
72
73
74
class ErrorResponse(OpenAIBaseModel):
@@ -1287,6 +1288,20 @@ class BatchRequestInput(OpenAIBaseModel):
1288
# The parameters of the request.
1289
body: Union[ChatCompletionRequest, EmbeddingRequest, ScoreRequest]
1290
1291
+ @field_validator('body', mode='plain')
1292
+ @classmethod
1293
+ def check_type_for_url(cls, value: Any, info: ValidationInfo):
1294
+ # Use url to disambiguate models
1295
+ url = info.data['url']
1296
+ if url == "/v1/chat/completions":
1297
+ return ChatCompletionRequest.model_validate(value)
1298
+ if url == "/v1/embeddings":
1299
+ return TypeAdapter(EmbeddingRequest).validate_python(value)
1300
+ if url == "/v1/score":
1301
+ return ScoreRequest.model_validate(value)
1302
+ return TypeAdapter(Union[ChatCompletionRequest, EmbeddingRequest,
1303
+ ScoreRequest]).validate_python(value)
1304
+
1305
1306
class BatchResponseData(OpenAIBaseModel):
1307
# HTTP status code of the response.