ZIBEE/bi_150-vllm
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Upgrade to vllm 0.17.0 corex v4.1 overlay

Browse Source
This commit is contained in:
LZiBee
2026-04-29 19:38:22 +08:00
parent 8fac6062e4
commit 938d0854a5
430 changed files with 35969 additions and 14511 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
vllm/entrypoints/openai/responses/protocol.py
View File

@@ -328,8 +328,9 @@ class ResponsesRequest(OpenAIBaseModel):
# Also check text.format for OpenAI-style json_schema
if self.text is not None and self.text.format is not None:
if structured_outputs is not None:
raise ValueError(
"Cannot specify both structured_outputs and text.format"
raise VLLMValidationError(
"Cannot specify both structured_outputs and text.format",
parameter="structured_outputs",
)
response_format = self.text.format
if (
@@ -378,14 +379,19 @@ class ResponsesRequest(OpenAIBaseModel):
)
@model_validator(mode="before")
@classmethod
def validate_background(cls, data):
if not data.get("background"):
return data
if not data.get("store", True):
raise ValueError("background can only be used when `store` is true")
raise VLLMValidationError(
"background can only be used when `store` is true",
parameter="background",
)
return data
@model_validator(mode="before")
@classmethod
def validate_prompt(cls, data):
if data.get("prompt") is not None:
raise VLLMValidationError(
@@ -394,16 +400,19 @@ class ResponsesRequest(OpenAIBaseModel):
return data
@model_validator(mode="before")
@classmethod
def check_cache_salt_support(cls, data):
if data.get("cache_salt") is not None and (
not isinstance(data["cache_salt"], str) or not data["cache_salt"]
):
raise ValueError(
"Parameter 'cache_salt' must be a non-empty string if provided."
raise VLLMValidationError(
"Parameter 'cache_salt' must be a non-empty string if provided.",
parameter="cache_salt",
)
return data
@model_validator(mode="before")
@classmethod
def function_call_parsing(cls, data):
"""Parse function_call dictionaries into ResponseFunctionToolCall objects.
This ensures Pydantic can properly resolve union types in the input field.

79
vllm/entrypoints/openai/responses/serving.py
View File

@@ -85,6 +85,8 @@ from vllm.entrypoints.openai.responses.protocol import (
ResponseCreatedEvent,
ResponseInProgressEvent,
ResponseInputOutputMessage,
ResponseReasoningPartAddedEvent,
ResponseReasoningPartDoneEvent,
ResponsesRequest,
ResponsesResponse,
ResponseUsage,
@@ -1339,6 +1341,19 @@ class OpenAIServingResponses(OpenAIServing):
),
)
)
yield _increment_sequence_number_and_return(
ResponseReasoningPartAddedEvent(
type="response.reasoning_part.added",
sequence_number=-1,
output_index=current_output_index,
item_id=current_item_id,
content_index=current_content_index,
part=ResponseReasoningTextContent(
text="",
type="reasoning_text",
),
)
)
else:
yield _increment_sequence_number_and_return(
ResponseOutputItemAddedEvent(
@@ -1354,22 +1369,21 @@ class OpenAIServingResponses(OpenAIServing):
),
)
)
yield _increment_sequence_number_and_return(
ResponseContentPartAddedEvent(
type="response.content_part.added",
sequence_number=-1,
output_index=current_output_index,
item_id=current_item_id,
content_index=current_content_index,
part=ResponseOutputText(
type="output_text",
text="",
annotations=[],
logprobs=[],
),
yield _increment_sequence_number_and_return(
ResponseContentPartAddedEvent(
type="response.content_part.added",
sequence_number=-1,
output_index=current_output_index,
item_id=current_item_id,
content_index=current_content_index,
part=ResponseOutputText(
type="output_text",
text="",
annotations=[],
logprobs=[],
),
)
)
)
current_content_index += 1
first_delta_sent = True
# todo(kebe7jun) tool call support
@@ -1397,6 +1411,19 @@ class OpenAIServingResponses(OpenAIServing):
text=reason_content,
)
)
yield _increment_sequence_number_and_return(
ResponseReasoningPartDoneEvent(
type="response.reasoning_part.done",
sequence_number=-1,
item_id=current_item_id,
output_index=current_output_index,
content_index=current_content_index,
part=ResponseReasoningTextContent(
text=reason_content,
type="reasoning_text",
),
)
)
current_content_index = 0
reasoning_item = ResponseReasoningItem(
type="reasoning",
@@ -1418,6 +1445,8 @@ class OpenAIServingResponses(OpenAIServing):
item=reasoning_item,
)
)
current_output_index += 1
current_item_id = str(uuid.uuid4())
yield _increment_sequence_number_and_return(
ResponseOutputItemAddedEvent(
type="response.output_item.added",
@@ -1432,8 +1461,6 @@ class OpenAIServingResponses(OpenAIServing):
),
)
)
current_output_index += 1
current_item_id = str(uuid.uuid4())
yield _increment_sequence_number_and_return(
ResponseContentPartAddedEvent(
type="response.content_part.added",
@@ -1449,7 +1476,6 @@ class OpenAIServingResponses(OpenAIServing):
),
)
)
current_content_index += 1
# reset previous delta messages
previous_delta_messages = []
@@ -1485,7 +1511,6 @@ class OpenAIServingResponses(OpenAIServing):
),
)
)
current_content_index += 1
previous_delta_messages.append(delta_message)
if previous_delta_messages:
@@ -1505,7 +1530,19 @@ class OpenAIServingResponses(OpenAIServing):
text=reason_content,
)
)
current_content_index += 1
yield _increment_sequence_number_and_return(
ResponseReasoningPartDoneEvent(
type="response.reasoning_part.done",
sequence_number=-1,
item_id=current_item_id,
output_index=current_output_index,
content_index=current_content_index,
part=ResponseReasoningTextContent(
text=reason_content,
type="reasoning_text",
),
)
)
reasoning_item = ResponseReasoningItem(
type="reasoning",
content=[
@@ -1543,7 +1580,6 @@ class OpenAIServingResponses(OpenAIServing):
item_id=current_item_id,
)
)
current_content_index += 1
part = ResponseOutputText(
text=final_content,
type="output_text",
@@ -1559,7 +1595,6 @@ class OpenAIServingResponses(OpenAIServing):
part=part,
)
)
current_content_index += 1
item = ResponseOutputMessage(
type="message",
role="assistant",