[Frontend][Bug] allow tool calls in analysis channel (#28139)
Signed-off-by: Marko Rosenmueller <5467316+dr75@users.noreply.github.com> Co-authored-by: Chauncey <chaunceyjiang@gmail.com>
This commit is contained in:
Marko Rosenmueller
committed by
GitHub
GitHub
parent
086b96339f
commit
455949675d
@@ -51,6 +51,9 @@ from vllm.entrypoints.openai.protocol import (
|
||||
ToolCall,
|
||||
UsageInfo,
|
||||
)
|
||||
from vllm.entrypoints.openai.serving_chat_stream_harmony import (
|
||||
extract_harmony_streaming_delta,
|
||||
)
|
||||
from vllm.entrypoints.openai.serving_engine import (
|
||||
GenerationError,
|
||||
OpenAIServing,
|
||||
@@ -837,64 +840,17 @@ class OpenAIServingChat(OpenAIServing):
|
||||
current_token_ids = as_list(output.token_ids)
|
||||
|
||||
if self.use_harmony:
|
||||
if cur_channel == "final":
|
||||
delta_message = DeltaMessage(content=delta_text)
|
||||
elif cur_channel == "analysis":
|
||||
if request.include_reasoning:
|
||||
delta_message = DeltaMessage(reasoning=delta_text)
|
||||
else:
|
||||
delta_message = None
|
||||
elif (
|
||||
cur_channel == "commentary"
|
||||
and cur_recipient
|
||||
and cur_recipient.startswith("functions.")
|
||||
):
|
||||
# Count completed tool calls to determine index
|
||||
base_index = 0
|
||||
for msg in harmony_parser.messages:
|
||||
if (
|
||||
msg.channel == "commentary"
|
||||
and msg.recipient
|
||||
and msg.recipient.startswith("functions.")
|
||||
):
|
||||
base_index += 1
|
||||
|
||||
if prev_recipient != cur_recipient:
|
||||
tool_name = cur_recipient.split("functions.", 1)[1]
|
||||
delta_message = DeltaMessage(
|
||||
tool_calls=[
|
||||
DeltaToolCall(
|
||||
id=make_tool_call_id(),
|
||||
type="function",
|
||||
function=DeltaFunctionCall(
|
||||
name=tool_name,
|
||||
arguments="",
|
||||
),
|
||||
index=base_index,
|
||||
)
|
||||
]
|
||||
)
|
||||
elif delta_text:
|
||||
delta_message = DeltaMessage(
|
||||
tool_calls=[
|
||||
DeltaToolCall(
|
||||
index=base_index,
|
||||
function=DeltaFunctionCall(
|
||||
arguments=delta_text
|
||||
),
|
||||
)
|
||||
]
|
||||
)
|
||||
else:
|
||||
delta_message = None
|
||||
|
||||
if delta_message is not None:
|
||||
harmony_tools_streamed[i] = True
|
||||
elif cur_channel == "commentary":
|
||||
# Tool call preambles meant to be shown to the user
|
||||
delta_message = DeltaMessage(content=delta_text)
|
||||
else:
|
||||
delta_message = None
|
||||
delta_message, tools_streamed_flag = (
|
||||
extract_harmony_streaming_delta(
|
||||
harmony_parser=harmony_parser,
|
||||
cur_channel=cur_channel,
|
||||
cur_recipient=cur_recipient,
|
||||
prev_recipient=prev_recipient,
|
||||
delta_text=delta_text,
|
||||
include_reasoning=request.include_reasoning,
|
||||
)
|
||||
)
|
||||
harmony_tools_streamed[i] |= tools_streamed_flag
|
||||
# handle streaming deltas for tools with named tool_choice
|
||||
elif tool_choice_function_name:
|
||||
if (
|
||||
|
||||
101
vllm/entrypoints/openai/serving_chat_stream_harmony.py
Normal file
101
vllm/entrypoints/openai/serving_chat_stream_harmony.py
Normal file
@@ -0,0 +1,101 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
"""
|
||||
Harmony-specific streaming delta extraction for chat completions.
|
||||
|
||||
This module handles the extraction of DeltaMessage objects from
|
||||
harmony parser state during streaming chat completions.
|
||||
"""
|
||||
|
||||
from openai_harmony import StreamableParser
|
||||
|
||||
from vllm.entrypoints.chat_utils import make_tool_call_id
|
||||
from vllm.entrypoints.openai.protocol import (
|
||||
DeltaFunctionCall,
|
||||
DeltaMessage,
|
||||
DeltaToolCall,
|
||||
)
|
||||
|
||||
|
||||
def extract_harmony_streaming_delta(
|
||||
harmony_parser: StreamableParser,
|
||||
cur_channel: str | None,
|
||||
cur_recipient: str | None,
|
||||
prev_recipient: str | None,
|
||||
delta_text: str,
|
||||
include_reasoning: bool,
|
||||
) -> tuple[DeltaMessage | None, bool]:
|
||||
"""
|
||||
Extract a DeltaMessage from harmony parser state during streaming.
|
||||
|
||||
Args:
|
||||
harmony_parser: The StreamableParser instance tracking parse state
|
||||
cur_channel: Current channel ("final", "analysis", "commentary", etc.)
|
||||
cur_recipient: Current recipient (e.g., "functions.my_func")
|
||||
prev_recipient: Previous recipient for detecting tool call transitions
|
||||
delta_text: The text delta to include in the message
|
||||
include_reasoning: Whether to include reasoning content
|
||||
|
||||
Returns:
|
||||
A tuple of (DeltaMessage or None, tools_streamed_flag)
|
||||
"""
|
||||
tools_streamed = False
|
||||
|
||||
if cur_channel == "final":
|
||||
delta_message = DeltaMessage(content=delta_text)
|
||||
elif (
|
||||
(cur_channel == "commentary" or cur_channel == "analysis")
|
||||
and cur_recipient
|
||||
and cur_recipient.startswith("functions.")
|
||||
):
|
||||
# Count completed tool calls to determine index
|
||||
base_index = 0
|
||||
for msg in harmony_parser.messages:
|
||||
if (
|
||||
(msg.channel == "commentary" or msg.channel == "analysis")
|
||||
and msg.recipient
|
||||
and msg.recipient.startswith("functions.")
|
||||
):
|
||||
base_index += 1
|
||||
|
||||
if prev_recipient != cur_recipient:
|
||||
tool_name = cur_recipient.split("functions.", 1)[1]
|
||||
delta_message = DeltaMessage(
|
||||
tool_calls=[
|
||||
DeltaToolCall(
|
||||
id=make_tool_call_id(),
|
||||
type="function",
|
||||
function=DeltaFunctionCall(
|
||||
name=tool_name,
|
||||
arguments="",
|
||||
),
|
||||
index=base_index,
|
||||
)
|
||||
]
|
||||
)
|
||||
elif delta_text:
|
||||
delta_message = DeltaMessage(
|
||||
tool_calls=[
|
||||
DeltaToolCall(
|
||||
index=base_index,
|
||||
function=DeltaFunctionCall(arguments=delta_text),
|
||||
)
|
||||
]
|
||||
)
|
||||
else:
|
||||
delta_message = None
|
||||
|
||||
if delta_message is not None:
|
||||
tools_streamed = True
|
||||
elif cur_channel == "commentary":
|
||||
# Tool call preambles meant to be shown to the user
|
||||
delta_message = DeltaMessage(content=delta_text)
|
||||
elif cur_channel == "analysis":
|
||||
if include_reasoning:
|
||||
delta_message = DeltaMessage(reasoning=delta_text)
|
||||
else:
|
||||
delta_message = None
|
||||
else:
|
||||
delta_message = None
|
||||
|
||||
return delta_message, tools_streamed
|
||||
Reference in New Issue
Block a user