biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

92
tests/entrypoints/openai/tool_parsers/utils.py
View File

@@ -4,15 +4,17 @@
from collections.abc import Iterable
from typing import Union
from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
DeltaMessage,
ExtractedToolCallInformation,
FunctionCall, ToolCall)
from vllm.entrypoints.openai.protocol import (
ChatCompletionRequest,
DeltaMessage,
ExtractedToolCallInformation,
FunctionCall,
ToolCall,
)
from vllm.entrypoints.openai.tool_parsers import ToolParser
class StreamingToolReconstructor:
def __init__(self, assert_one_tool_per_delta: bool = True):
self.tool_calls: list[ToolCall] = []
self.other_content: str = ""
@@ -23,49 +25,60 @@ class StreamingToolReconstructor:
self.other_content += delta.content
else:
assert delta.tool_calls, (
"Streaming results should have either content or tool calls "
"(or both)")
"Streaming results should have either content or tool calls (or both)"
)
if self._assert_one_tool_per_delta:
# Note: This isn't strictly required by the API and may not be
# possible to adhere to depending on the token space and number of
# tokens per streamed response from the model, but it is required
# by tool_use tests, so we enforce it here by default also.
assert len(delta.tool_calls) < 2, (
"Streaming should include only one tool call per update.")
"Streaming should include only one tool call per update."
)
for call_delta in delta.tool_calls:
assert call_delta.type is None or call_delta.type == "function", (
"Streaming tool calls should only emit function calls. Got "
f"{call_delta.type}")
current_tool_call = self.tool_calls[
call_delta.index] if call_delta.index < len(
self.tool_calls) else None
f"{call_delta.type}"
)
current_tool_call = (
self.tool_calls[call_delta.index]
if call_delta.index < len(self.tool_calls)
else None
)
if current_tool_call:
assert (not call_delta.function.name), (
assert not call_delta.function.name, (
"Streaming tool calls should emit the full function name "
f"exactly once. Got {call_delta.function.name}")
assert (not call_delta.id), (
f"exactly once. Got {call_delta.function.name}"
)
assert not call_delta.id, (
"Streaming tool calls must emit function id only once. Got "
f"{call_delta.id}")
assert (call_delta.index == len(self.tool_calls) - 1), (
f"{call_delta.id}"
)
assert call_delta.index == len(self.tool_calls) - 1, (
f"Incorrect index for tool delta. Got {call_delta.index}, "
f"expected {len(self.tool_calls) - 1}")
current_tool_call.function.arguments += (
call_delta.function.arguments)
f"expected {len(self.tool_calls) - 1}"
)
current_tool_call.function.arguments += call_delta.function.arguments
else:
assert call_delta.id is not None, (
"Streaming tool calls must have an id on first appearance")
"Streaming tool calls must have an id on first appearance"
)
assert call_delta.function.name is not None, (
"Streaming tool calls must have a function name on first "
"appearance")
"Streaming tool calls must have a function name on first appearance"
)
assert call_delta.index == len(self.tool_calls), (
f"Incorrect index for tool delta. Got {call_delta.index}, "
f"expected {len(self.tool_calls)}")
f"expected {len(self.tool_calls)}"
)
self.tool_calls.append(
ToolCall(id=call_delta.id,
function=FunctionCall(
name=call_delta.function.name,
arguments=call_delta.function.arguments
or "")))
ToolCall(
id=call_delta.id,
function=FunctionCall(
name=call_delta.function.name,
arguments=call_delta.function.arguments or "",
),
)
)
def run_tool_extraction(
@@ -80,11 +93,11 @@ def run_tool_extraction(
tool_parser,
model_output,
request,
assert_one_tool_per_delta=assert_one_tool_per_delta)
assert_one_tool_per_delta=assert_one_tool_per_delta,
)
return reconstructor.other_content or None, reconstructor.tool_calls
else:
extracted = run_tool_extraction_nonstreaming(tool_parser, model_output,
request)
extracted = run_tool_extraction_nonstreaming(tool_parser, model_output, request)
assert extracted.tools_called == bool(extracted.tool_calls)
return extracted.content, extracted.tool_calls
@@ -92,7 +105,7 @@ def run_tool_extraction(
def run_tool_extraction_nonstreaming(
tool_parser: ToolParser,
model_output: str,
request: Union[ChatCompletionRequest, None] = None
request: Union[ChatCompletionRequest, None] = None,
) -> ExtractedToolCallInformation:
request = request or ChatCompletionRequest(messages=[], model="test-model")
return tool_parser.extract_tool_calls(model_output, request)
@@ -106,7 +119,8 @@ def run_tool_extraction_streaming(
) -> StreamingToolReconstructor:
request = request or ChatCompletionRequest(messages=[], model="test-model")
reconstructor = StreamingToolReconstructor(
assert_one_tool_per_delta=assert_one_tool_per_delta)
assert_one_tool_per_delta=assert_one_tool_per_delta
)
previous_text = ""
previous_tokens: list[int] = []
for delta in model_deltas:
@@ -118,8 +132,14 @@ def run_tool_extraction_streaming(
current_text = previous_text + delta
current_tokens = previous_tokens + token_delta
delta_message = tool_parser.extract_tool_calls_streaming(
previous_text, current_text, delta, previous_tokens,
current_tokens, token_delta, request)
previous_text,
current_text,
delta,
previous_tokens,
current_tokens,
token_delta,
request,
)
if delta_message is not None:
reconstructor.append_delta(delta_message)
previous_text = current_text