Skip to content

Conversation

@wenmengzhou
Copy link

@wenmengzhou wenmengzhou commented Dec 2, 2025

Purpose

add tool parser support for deepseek v3.2

Test Plan

linter passed
add UT passed

Test Result

============================================================================================================================================== test session starts ==============================================================================================================================================
platform darwin -- Python 3.10.14, pytest-8.3.5, pluggy-1.5.0
rootdir: /Users/zhouwenmeng/work/code/llm/vllm
configfile: pyproject.toml
plugins: anyio-4.9.0, asyncio-0.25.3, cov-6.2.1, langsmith-0.3.13, hydra-core-1.3.2
asyncio: mode=strict, asyncio_default_fixture_loop_scope=None
collected 3 items                                                                                                                                                                                                                                                                                               

tests/tool_use/test_deepseek_v32_tool_parser.py ...                                                                                                                                                                                                                                                       [100%]

=============================================================================================================================================== warnings summary ================================================================================================================================================
vllm/__init__.py:7
  /Users/zhouwenmeng/work/code/llm/vllm/vllm/__init__.py:7: RuntimeWarning: Failed to read commit hash:
  No module named 'vllm._version'
    from .version import __version__, __version_tuple__  # isort:skip

<frozen importlib._bootstrap>:241
  <frozen importlib._bootstrap>:241: DeprecationWarning: builtin type SwigPyPacked has no __module__ attribute

<frozen importlib._bootstrap>:241
  <frozen importlib._bootstrap>:241: DeprecationWarning: builtin type SwigPyObject has no __module__ attribute

vllm/entrypoints/openai/speech_to_text.py:40
  /Users/zhouwenmeng/work/code/llm/vllm/vllm/entrypoints/openai/speech_to_text.py:40: DeprecationWarning: `vllm.transformers_utils.tokenizer.get_tokenizer` has been moved to `vllm.tokenizers.get_tokenizer`. The old name will be removed in v0.13.
    from vllm.transformers_utils.tokenizer import get_tokenizer

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
========================================================================================================================================= 3 passed, 4 warnings in 5.09s =========================================================================================================================================

Essential Elements of an Effective PR Description Checklist
  • [ x] add support for dpsk ".
  • [x ] The test plan, such as providing test command.
  • [x ] The test results, such as pasting the results comparison before and after, or e2e results

@mergify mergify bot added deepseek Related to DeepSeek models frontend qwen Related to Qwen models tool-calling labels Dec 2, 2025
Copy link

@chatgpt-codex-connector chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

Comment on lines 392 to 396
if (
buffer.startswith("<DSML_function_calls")
or buffer == "<DSML_function_calls"[: len(buffer)]
or buffer.startswith("<DSML_invoke")
or buffer == "<DSML_invoke"[: len(buffer)]

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Handle partial DeepSeek tags with original token prefixes

When the stream delivers a split token such as <|DSML|function_calls without the closing >, _find_next_complete_element only waits if the buffer starts with <DSML_…> and otherwise treats the fragment as plain text. DeepSeek actually emits <|DSML|…> prefixes, so any tag broken across chunks will be consumed as text and the tool call never parsed, breaking streaming extraction whenever tags are split mid-token.

Useful? React with 👍 / 👎.

Comment on lines +1141 to +1144
logger.warning(
"Parsed value '%s' of parameter '%s' is not an integer "
"in tool '%s', degenerating to string.",
param_value,

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Fix logging argument mismatch in numeric parse fallback

The warning emitted when integer parsing fails supplies only param_value to a format string with three %s placeholders, so hitting this path raises TypeError inside logging and aborts parsing instead of gracefully falling back to a string. The same mismatch appears in the float branch below; any non-numeric output for numeric parameters will crash the parser rather than returning the raw value.

Useful? React with 👍 / 👎.

Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request introduces a tool parser for DeepSeek v3.2 by adapting the existing Qwen XML parser. While the implementation is comprehensive for both streaming and non-streaming scenarios, I have significant concerns about the robustness and maintainability of the core parsing logic. The StreamingXMLToolCallParser contains a critical flaw in its stream-splitting mechanism that can lead to parser failures on valid tool calls. Furthermore, the overall complexity of this class is very high, making it difficult to maintain. The accompanying tests also lack coverage for crucial edge cases that could expose these fragilities. My review includes detailed feedback on these critical and high-severity issues with recommendations for simplification and improved testing.

Comment on lines 355 to 418
def _find_next_complete_element(self, start_pos: int) -> tuple[str | None, int]:
"""
Find next complete XML element from specified position
Args:
start_pos: Position to start searching
Returns:
(Complete element string, element end position),
returns (None, start_pos) if no complete element found
"""
buffer = self.streaming_buffer[start_pos:]

if not buffer:
return None, start_pos

if buffer.startswith("<"):
# Need to ensure no new < appears,
# find the nearest one between < and >
tag_end = buffer.find("<", 1)
tag_end2 = buffer.find(">", 1)
if tag_end != -1 and tag_end2 != -1:
# Next nearest is <
if tag_end < tag_end2:
return buffer[:tag_end], start_pos + tag_end
# Next nearest is >, means found XML element
else:
return buffer[: tag_end2 + 1], start_pos + tag_end2 + 1
elif tag_end != -1:
return buffer[:tag_end], start_pos + tag_end
elif tag_end2 != -1:
return buffer[: tag_end2 + 1], start_pos + tag_end2 + 1
else:
# If currently not parsing tool calls (entering a tool_call),
# check if starts with DeepSeek tags
if self.current_call_id is None:
# Check if might be start of DeepSeek tags
if (
buffer.startswith("<DSML_function_calls")
or buffer == "<DSML_function_calls"[: len(buffer)]
or buffer.startswith("<DSML_invoke")
or buffer == "<DSML_invoke"[: len(buffer)]
):
# Might be start of DeepSeek tag, wait for more data
return None, start_pos
else:
# Not start of DeepSeek tag, treat as text
return buffer, start_pos + len(buffer)
else:
# When parsing tool calls,
# wait for more data to get complete tag
return None, start_pos
else:
# Find text content (until next < or buffer end)
next_tag_pos = buffer.find("<")
if next_tag_pos != -1:
# Found text content
text_content = buffer[:next_tag_pos]
return text_content, start_pos + next_tag_pos
else:
# Buffer end is all text, process
# (no longer wait for more data)
remaining = buffer
return remaining, start_pos + len(remaining)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

critical

The function _find_next_complete_element has a fundamental flaw in its parsing logic. It splits the input stream by finding the next < or > character. This approach does not account for < or > characters appearing inside quoted attribute values, which is valid in XML. For example, an input like <invoke name="a<b"> would be incorrectly split into <invoke name="a<, which is not a valid tag and will cause the expat parser to fail with an "unclosed token" error. This makes the entire tool call parsing mechanism fragile and prone to errors with certain model outputs.

A more robust approach should be used. Instead of manually trying to find complete XML elements, I recommend simplifying the parsing logic to leverage expat's streaming capabilities more directly. The expat parser is designed to handle chunked data. The pre-processing should be limited to what's necessary, like replacing DeepSeek's special tokens.

If the goal is to sanitize the input before it reaches expat, a more sophisticated state machine that is aware of XML quoting rules is needed for splitting the stream. However, this adds a lot of complexity, and it's usually better to rely on the battle-tested expat parser as much as possible.

Comment on lines 31 to 1261
class StreamingXMLToolCallParser:
"""
Simplified streaming XML tool call parser
Supports streaming input, parsing, and output
"""

def __init__(self):
self.reset_streaming_state()

# Tool configuration information
self.tools: list[ChatCompletionToolsParam] | None = None

# DeepSeek v3.2 token definitions
self.tool_call_start_token: str = "<|DSML|function_calls>"
self.tool_call_end_token: str = "</|DSML|function_calls>"
self.function_start_token: str = "<|DSML|invoke"
self.function_end_token: str = "</|DSML|invoke>"
self.parameter_start_token: str = "<|DSML|parameter"
self.parameter_end_token: str = "</|DSML|parameter>"
self.end_of_sentence_token: str = "<|end▁of▁sentence|>"

def reset_streaming_state(self):
"""Reset streaming parsing state"""

self.deltas = []
# state for streaming
self.tool_call_index = 0
self.current_call_id = None
self.last_completed_call_id = None
self.current_function_name = None
self.current_function_open = False
self.parameters = {}
self.current_param_name = None
self.current_param_value = ""
self.current_param_value_converted = ""
self.current_param_is_first = False
self.should_emit_end_newline = False
self.start_quote_emitted = False

self.streaming_buffer = ""
self.last_processed_pos = 0

self.text_content_buffer = ""

# state for preprocessing and deferred parsing
self._pre_inside_parameter = False
self._pre_param_buffer = ""
self._pre_current_param_name = None
self.defer_current_parameter = False
self.deferred_param_raw_value = ""

# DeepSeek explicit type support
self.current_param_explicit_type = None

# recreate parser
self.parser = ParserCreate()
self.setup_parser()

def parse_single_streaming_chunks(self, xml_chunk: str) -> DeltaMessage:
"""
Parse single streaming XML chunk and return Delta response
This is the actual streaming interface that receives chunks
one by one and maintains internal state
Args:
xml_chunk: Single XML chunk string
Returns:
DeltaMessage: Contains delta information generated by this chunk,
returns empty response if no complete elements
"""
# Remove DeepSeek end-of-sentence token if present
if self.end_of_sentence_token and self.end_of_sentence_token in xml_chunk:
xml_chunk = xml_chunk.replace(self.end_of_sentence_token, "")

# Record delta count before processing
initial_delta_count = len(self.deltas)

self.streaming_buffer += xml_chunk

found_elements = self._process_complete_xml_elements()

if found_elements:
# If complete elements found, check if end events were missed
# some tags may not have been triggered
try:
new_deltas = self.deltas[initial_delta_count:]
# If this chunk contains </|DSML|invoke>
# but didn't generate '}', then complete it
if (
self.current_call_id is not None
and "</|DSML|invoke>" in xml_chunk
):
# - Added '}' (non-empty parameter ending)
# - Added '{}' (empty parameter function)
has_function_close = any(
(
td.tool_calls
and any(
(
tc.function
and tc.id == self.current_call_id
and isinstance(tc.function.arguments, str)
and (tc.function.arguments in ("}", "{}"))
)
for tc in td.tool_calls
)
)
for td in new_deltas
)
if not has_function_close:
# Close potentially unclosed element
if self.current_param_name:
self._end_element("DSML_parameter")
if self.current_function_name:
self._end_element("DSML_invoke")
# If this chunk contains </|DSML|function_calls>
# but didn't generate final empty delta, then complete it
if (
self.current_call_id is not None
and "</|DSML|function_calls>" in xml_chunk
):
has_container_close = any(
(
td.tool_calls
and any(
(
tc.type == "function"
and tc.function
and tc.function.arguments == ""
and tc.id == self.current_call_id
)
for tc in td.tool_calls
)
)
for td in new_deltas
)
if not has_container_close:
# Close potentially unclosed element
if self.current_param_name:
self._end_element("DSML_parameter")
if self.current_function_name:
self._end_element("DSML_invoke")
self._end_element("DSML_function_calls")
except Exception as e:
logger.warning("Error with fallback parsing: %s", e)
# Merge newly generated deltas into single response
result_delta = self._merge_new_deltas_to_single_response(
initial_delta_count
)
return result_delta
else:
# No complete elements, check if there's unoutput text content
if self.text_content_buffer and self.tool_call_index == 0:
# Has text content but no tool_call yet, output text content
text_delta = DeltaMessage(content=self.text_content_buffer)
self._emit_delta(text_delta)
# Clear buffer to avoid duplicate output
self.text_content_buffer = ""
return text_delta

# If this chunk contains end tags but wasn't triggered by parser,
# manually complete end events
# Only execute when still on the same call as when entered,
# to prevent accidentally closing new calls
# in multi invoke scenarios
if self.current_call_id is not None and (
"</|DSML|invoke>" in xml_chunk
or "</|DSML|function_calls>" in xml_chunk
):
# Close potentially unclosed element
if self.current_param_name:
self._end_element("DSML_parameter")
if "</|DSML|invoke>" in xml_chunk and self.current_function_name:
self._end_element("DSML_invoke")
if "</|DSML|function_calls>" in xml_chunk:
self._end_element("DSML_function_calls")
# Return the merged delta result generated by this fallback
result_delta = self._merge_new_deltas_to_single_response(
initial_delta_count
)
return result_delta

# No complete elements, return empty response
return DeltaMessage(content=None)

def _escape_xml_special_chars(self, text: str) -> str:
"""
Escape XML special characters
Args:
text: Original text
Returns:
Escaped text
"""
xml_escapes = {
"&": "&amp;",
"<": "&lt;",
">": "&gt;",
'"': "&quot;",
"'": "&apos;",
}

for char, escape in xml_escapes.items():
text = text.replace(char, escape)

return text

def _process_complete_xml_elements(self) -> bool:
"""
Process complete XML elements in buffer
Returns:
bool: Whether complete elements were found and processed
"""
found_any = False
while self.last_processed_pos < len(self.streaming_buffer):
# Find next complete xml element
element, end_pos = self._find_next_complete_element(self.last_processed_pos)
if element is None:
# No complete element found, wait for more data
break

# Check if this element should be skipped
if self._should_skip_element(element):
self.last_processed_pos = end_pos
continue

# Found complete XML element, process it
try:
preprocessed_element = self._preprocess_xml_chunk(element)
# Check if this is the first tool_call start
# Note: after preprocessing, <|DSML|invoke becomes <DSML_invoke
if (
(
preprocessed_element.strip().startswith("<DSML_invoke")
or preprocessed_element.strip().startswith(
"<DSML_function_calls"
)
)
and self.tool_call_index == 0
) and self.text_content_buffer:
# First tool_call starts,
# output previously collected text content first
text_delta = DeltaMessage(content=self.text_content_buffer)
self._emit_delta(text_delta)
# Clear buffer for potential subsequent text content
self.text_content_buffer = ""

# If a new invoke starts and
# there are already completed invokes
if (
preprocessed_element.strip().startswith("<DSML_invoke")
and self.tool_call_index > 0
and self.current_call_id
):
# Reset parser state but preserve generated deltas
if self.current_param_name:
self._end_element("DSML_parameter")
if self.current_function_open or self.current_function_name:
self._end_element("DSML_invoke")
# Output final invoke tail delta
final_delta = DeltaMessage(
role=None,
content=None,
reasoning=None,
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments=""),
)
],
)
self._emit_delta(final_delta)
# Reset current call state (lightweight reset for DeepSeek)
self._reset_current_call_state()
# Parse preprocessed element
self.parser.Parse(preprocessed_element, False)
found_any = True

except Exception as e:
logger.warning("Error when parsing XML elements: %s", e)

# Update processed position
self.last_processed_pos = end_pos

return found_any

def _should_skip_element(self, element: str) -> bool:
"""
Determine whether an element should be skipped
Args:
element: Element to evaluate
Returns:
bool: True means should skip, False means should process
"""

# Check if it's a DeepSeek tool_call XML tag, don't skip
if (
element.startswith(self.tool_call_start_token)
or element.startswith(self.function_start_token)
or element.startswith(self.parameter_start_token)
):
return False

# If currently not parsing tool calls and not blank,
# collect this text instead of skipping
# Only process other XML elements after tool_call appears,
# otherwise treat as plain text
if self.current_call_id is None and element:
# Collect text content to buffer
self.text_content_buffer += element
return True # Still skip, but content has been collected

# If currently parsing tool calls,
# this might be parameter value, don't skip
if self.current_call_id is not None:
return False

# Skip blank content
return not element

def _find_next_complete_element(self, start_pos: int) -> tuple[str | None, int]:
"""
Find next complete XML element from specified position
Args:
start_pos: Position to start searching
Returns:
(Complete element string, element end position),
returns (None, start_pos) if no complete element found
"""
buffer = self.streaming_buffer[start_pos:]

if not buffer:
return None, start_pos

if buffer.startswith("<"):
# Need to ensure no new < appears,
# find the nearest one between < and >
tag_end = buffer.find("<", 1)
tag_end2 = buffer.find(">", 1)
if tag_end != -1 and tag_end2 != -1:
# Next nearest is <
if tag_end < tag_end2:
return buffer[:tag_end], start_pos + tag_end
# Next nearest is >, means found XML element
else:
return buffer[: tag_end2 + 1], start_pos + tag_end2 + 1
elif tag_end != -1:
return buffer[:tag_end], start_pos + tag_end
elif tag_end2 != -1:
return buffer[: tag_end2 + 1], start_pos + tag_end2 + 1
else:
# If currently not parsing tool calls (entering a tool_call),
# check if starts with DeepSeek tags
if self.current_call_id is None:
# Check if might be start of DeepSeek tags
if (
buffer.startswith("<DSML_function_calls")
or buffer == "<DSML_function_calls"[: len(buffer)]
or buffer.startswith("<DSML_invoke")
or buffer == "<DSML_invoke"[: len(buffer)]
):
# Might be start of DeepSeek tag, wait for more data
return None, start_pos
else:
# Not start of DeepSeek tag, treat as text
return buffer, start_pos + len(buffer)
else:
# When parsing tool calls,
# wait for more data to get complete tag
return None, start_pos
else:
# Find text content (until next < or buffer end)
next_tag_pos = buffer.find("<")
if next_tag_pos != -1:
# Found text content
text_content = buffer[:next_tag_pos]
return text_content, start_pos + next_tag_pos
else:
# Buffer end is all text, process
# (no longer wait for more data)
remaining = buffer
return remaining, start_pos + len(remaining)

def _merge_new_deltas_to_single_response(self, initial_count: int) -> DeltaMessage:
"""
Merge newly generated deltas from this processing
into a single DeltaMessage
Args:
initial_count: Delta count before processing
Returns:
Merged DeltaMessage containing all newly generated delta information
"""
if len(self.deltas) <= initial_count:
return DeltaMessage(content=None)

# Get newly generated deltas
new_deltas = self.deltas[initial_count:]

if len(new_deltas) == 1:
# Only one new delta, return directly
return new_deltas[0]

# Merge multiple new deltas
merged_tool_calls: list[DeltaToolCall] = []
merged_content: str = ""

for delta in new_deltas:
if delta.content:
merged_content += delta.content
if delta.tool_calls:
# For tool_calls, we need to intelligently merge arguments
for tool_call in delta.tool_calls:
# Find if there's already a tool_call with the same call_id
existing_call = None
for existing in merged_tool_calls:
if existing.id == tool_call.id:
existing_call = existing
break

if existing_call and existing_call.function:
# Merge to existing tool_call
if tool_call.function and tool_call.function.name:
existing_call.function.name = tool_call.function.name
if (
tool_call.function
and tool_call.function.arguments is not None
):
if existing_call.function.arguments is None:
existing_call.function.arguments = ""

# For streaming JSON parameters,
# simply concatenate in order
new_args = tool_call.function.arguments
existing_call.function.arguments += new_args
if tool_call.type:
existing_call.type = tool_call.type
else:
# Add new tool_call
merged_tool_calls.append(tool_call)

return DeltaMessage(
content=merged_content if merged_content else None,
tool_calls=merged_tool_calls,
)

def _preprocess_xml_chunk(self, chunk: str) -> str:
"""
Preprocess XML chunk, handle DeepSeek special characters,
and escape special characters
Args:
chunk: Original XML chunk
Returns:
Processed XML chunk
"""

# Step 1: Normalize DeepSeek special characters to XML-safe characters
processed = chunk.replace("|DSML|", "DSML_")

# Check if this is a tool_call related element
is_tool_call = False
if (
"DSML_function_calls" in processed
or "DSML_invoke" in processed
or "DSML_parameter" in processed
):
is_tool_call = True

# DeepSeek already uses standard attribute format, no conversion needed
# <DSML_invoke name="get_weather"> ✓
# <DSML_parameter name="location" string="true"> ✓

original_chunk = chunk
# If in parameter value accumulation mode
if self._pre_inside_parameter:
# Parameter end: output accumulated raw text
# safely then return </parameter>
if processed.startswith("</DSML_parameter>"):
body_text = self._pre_param_buffer
# Trigger deferred parsing mode
# literal_eval+json output in end_element
self.defer_current_parameter = True
self.deferred_param_raw_value = body_text
# Clean up state
self._pre_inside_parameter = False
self._pre_param_buffer = ""
self._pre_current_param_name = None
safe_text = self._escape_xml_special_chars(body_text)
return f"{safe_text}</DSML_parameter>"
else:
# If this is the first block of content after entering parameter
# evaluate if deferred parsing is needed;
# If not needed, exit accumulation mode
# and pass through directly
if self._pre_param_buffer == "":
# Get current parameter type
param_type = (
self._get_param_type(self._pre_current_param_name)
if self._pre_current_param_name
else "string"
)
# Only these types need deferred parsing to
# handle Python literals containing single quotes
is_object_type = param_type in ["object"]
is_complex_type = (
param_type in ["array", "arr", "sequence"]
or param_type.startswith("dict")
or param_type.startswith("list")
)

# Only delay when contains container symbols
# and has single quotes and is complex type
has_container_hint = (
("[" in original_chunk)
or ("{" in original_chunk)
or ("(" in original_chunk)
)

# Determine if deferred parsing is needed
need_defer = False
if is_complex_type:
# Complex type, always need deferred parsing
need_defer = True
elif (
is_object_type
and has_container_hint
and ("'" in original_chunk)
):
# Object type with container symbols
# and single quotes, need deferred parsing
need_defer = True

if not need_defer:
# No need for deferred parsing,
# exit parameter mode directly
self._pre_inside_parameter = False
return self._escape_xml_special_chars(original_chunk)
self._pre_param_buffer += original_chunk
return ""

# Parameter start: enable accumulation
if processed.startswith("<DSML_parameter"):
# Extract parameter name and type attributes
m = re.match(
r'<DSML_parameter\s+name="([^"]+)"(?:\s+string="(true|false)")?(?:\s+number="(true|false)")?(?:\s+boolean="(true|false)")?(?:\s+object="(true|false)")?(?:\s+array="(true|false)")?',
processed,
)
if m:
self._pre_current_param_name = m.group(1)
self._pre_inside_parameter = True
self._pre_param_buffer = ""
return processed

# If processed doesn't contain special_token, escape processed
# This is because XML parsing encounters special characters
# and reports errors, so escaping is needed
if not is_tool_call:
processed = self._escape_xml_special_chars(processed)
return processed

def _emit_delta(self, delta: DeltaMessage):
"""Emit Delta response (streaming output)"""
self.deltas.append(delta)

def _auto_close_open_parameter_if_needed(self, incoming_tag: str | None = None):
"""Before starting to process new elements,
if there are unclosed tags from before,
automatically complete their endings to the parser.
- If there are unclosed parameters,
it's equivalent to feeding `</parameter>`
- When about to start a new function or tool_call,
if there are unclosed functions, complete `</function>`.
- When about to start a new tool_call,
if there are unclosed tool_calls, complete `</tool_call>`.
"""
# First close unclosed parameters
if self.current_param_name:
self._end_element("parameter")

# If about to start new function or tool_call,
# and there are unclosed functions, close function first
if incoming_tag in ("function", "tool_call") and self.current_function_name:
self._end_element("function")

# If about to start new tool_call,
# and there are unclosed tool_calls, close tool_call first
if incoming_tag == "tool_call" and self.current_call_id:
self._end_element("tool_call")

def _start_element(self, name: str, attrs: dict[str, str]):
"""Handle XML start element events for DeepSeek format"""

if name == "root":
return

# Handle function calls
if name == "DSML_function_calls":
# Before opening new tool_call,
# automatically complete previous unclosed tags
self._auto_close_open_parameter_if_needed("DSML_function_calls")

self.parameters = {}
self.current_call_id = make_tool_call_id()
self.current_param_is_first = True
self.tool_call_index += 1

# Handle invoke (equivalent to Qwen3's tool_call + function combined)
if name == "DSML_invoke":
# If missing tool_call, manually complete
if not self.current_call_id:
self._start_element("DSML_function_calls", {})

# Extract function name from invoke's name attribute
function_name = attrs.get("name")
if function_name:
self.current_function_name = function_name
self.current_function_open = True

# Emit delta with function name
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(
name=function_name, arguments=""
),
)
]
)
self._emit_delta(delta)

# Handle parameter
elif name == "DSML_parameter":
# If no invoke yet, create one (fault tolerance)
if not self.current_call_id:
self._start_element("DSML_invoke", {"name": "unknown"})

# Auto-close previous parameter if exists
self._auto_close_open_parameter_if_needed("DSML_parameter")

# Extract parameter name and type
param_name = attrs.get("name")
self.current_param_name = param_name
self.current_param_value = ""
self.current_param_value_converted = ""
self.start_quote_emitted = False

# Save explicit type information
if attrs.get("string") == "true":
self.current_param_explicit_type = "string"
elif attrs.get("number") == "true":
self.current_param_explicit_type = "number"
elif attrs.get("boolean") == "true":
self.current_param_explicit_type = "boolean"
elif attrs.get("object") == "true":
self.current_param_explicit_type = "object"
elif attrs.get("array") == "true":
self.current_param_explicit_type = "array"
else:
self.current_param_explicit_type = None

# Output JSON parameter name and colon
if param_name:
if not self.parameters:
# First parameter - start JSON object
json_start = f'{{"{param_name}": '
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(
name=None, arguments=json_start
),
)
]
)
self._emit_delta(delta)
self.current_param_is_first = True
else:
# Subsequent parameters - add comma
json_continue = f', "{param_name}": '
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(
name=None, arguments=json_continue
),
)
]
)
self._emit_delta(delta)
self.current_param_is_first = False

def _char_data(self, data: str):
"""Handle XML character data events"""
if data and self.current_param_name:
# If preprocessing stage determines deferred parsing is needed,
# only cache character data, no streaming output
if self.defer_current_parameter:
original_data = data
if self.should_emit_end_newline:
original_data = "\n" + original_data
self.should_emit_end_newline = False
if original_data.endswith("\n"):
self.should_emit_end_newline = True
original_data = original_data[:-1]
self.current_param_value += original_data
return

param_type = self._get_effective_param_type(self.current_param_name)

# Check if this is the first time receiving data for this parameter
# If this is the first packet of data and starts with \n, remove \n
if not self.current_param_value and data.startswith("\n"):
data = data[1:]

# Output start quote for string type (if not already output)
if (
param_type in ["string", "str", "text", "varchar", "char", "enum"]
and not self.start_quote_emitted
):
quote_delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments='"'),
)
]
)
self._emit_delta(quote_delta)
self.start_quote_emitted = True

if not data:
return

original_data = data
# Delay output of trailing newline
if self.should_emit_end_newline:
original_data = "\n" + original_data
self.should_emit_end_newline = False
if original_data.endswith("\n"):
self.should_emit_end_newline = True
original_data = original_data[:-1]
self.current_param_value += original_data

# convert parameter value by param_type
converted_value = self._convert_param_value(
self.current_param_value, param_type
)
output_data = self._convert_for_json_streaming(converted_value, param_type)

delta_data = output_data[len(self.current_param_value_converted) :]
self.current_param_value_converted = output_data

delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments=delta_data),
)
]
)
self._emit_delta(delta)

def _end_element(self, name: str):
"""Handle XML end element events for DeepSeek format"""

if name == "root":
return

# If invoke ends and there are still unclosed parameters,
# complete parameter end first
if (
name == "DSML_invoke" or name == "DSML_function_calls"
) and self.current_param_name:
self._auto_close_open_parameter_if_needed()

# Handle parameter ending
if name == "DSML_parameter" and self.current_param_name:
# End current parameter
param_name = self.current_param_name
param_value = self.current_param_value

# If in deferred parsing mode,
# perform overall parsing on raw content
# accumulated in preprocessing stage and output once
if self.defer_current_parameter:
raw_text = (
self.deferred_param_raw_value
if self.deferred_param_raw_value
else param_value
)
parsed_value = None
output_arguments = None
try:
# If previously delayed trailing newline,
# add it back before parsing
if self.should_emit_end_newline:
raw_for_parse = raw_text + "\n"
else:
raw_for_parse = raw_text
parsed_value = ast.literal_eval(raw_for_parse)
output_arguments = json.dumps(parsed_value, ensure_ascii=False)
except Exception:
# Fallback: output as string as-is
output_arguments = json.dumps(raw_text, ensure_ascii=False)
parsed_value = raw_text

delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(
name=None, arguments=output_arguments
),
)
]
)
self._emit_delta(delta)

# Clean up and store
self.should_emit_end_newline = False
self.parameters[param_name] = parsed_value
self.current_param_name = None
self.current_param_value = ""
self.current_param_value_converted = ""
self.start_quote_emitted = False
self.defer_current_parameter = False
self.deferred_param_raw_value = ""
# Reset explicit type
self.current_param_explicit_type = None
return

param_type = self._get_effective_param_type(param_name)

# convert complete parameter value by param_type
converted_value = self._convert_param_value(param_value, param_type)

# Decide whether to add end quote based on parameter type
if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
# For empty string parameters, need special handling
if not param_value and not self.start_quote_emitted:
# No start quote output,
# directly output complete empty string
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments='""'),
)
]
)
self._emit_delta(delta)
else:
# Non-empty parameter value, output end quote
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments='"'),
)
]
)
self._emit_delta(delta)

self.should_emit_end_newline = False
# Store converted value
self.parameters[param_name] = converted_value
self.current_param_name = None
self.current_param_value = ""
self.current_param_value_converted = ""
self.start_quote_emitted = False
# Reset explicit type
self.current_param_explicit_type = None

# Handle invoke ending (equivalent to Qwen3's function + tool_call)
elif name == "DSML_invoke":
# Ensure parameter is closed first
if self.current_param_name:
self._end_element("DSML_parameter")

# Close JSON object
if self.parameters:
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments="}"),
)
]
)
self._emit_delta(delta)
else:
# Empty parameters - output {}
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments="{}"),
)
]
)
self._emit_delta(delta)
self.current_function_open = False

elif name == "DSML_function_calls":
# Before ending tool_call,
# ensure function is closed to complete missing right brace
if self.current_function_open:
# If there are still unclosed parameters, close them first
if self.current_param_name:
self._end_element("DSML_parameter")
# Close function, ensure output '}' or '{}'
self._end_element("DSML_invoke")
delta = DeltaMessage(
tool_calls=[
DeltaToolCall(
index=self.tool_call_index - 1,
id=self.current_call_id,
type="function",
function=DeltaFunctionCall(name=None, arguments=""),
)
]
)
self._emit_delta(delta)

# Check if there's text content to output (between invokes)
if self.text_content_buffer.strip():
text_delta = DeltaMessage(content=self.text_content_buffer)
self._emit_delta(text_delta)

# Lightweight reset for next invoke (don't reset parser)
self._reset_current_call_state()

def setup_parser(self):
"""Set up XML parser event handlers"""
self.parser.buffer_text = True
self.parser.StartElementHandler = self._start_element
self.parser.EndElementHandler = self._end_element
self.parser.CharacterDataHandler = self._char_data

def set_tools(self, tools: list[ChatCompletionToolsParam] | None):
"""Set tool configuration information"""
self.tools = tools

def _extract_function_name(self, name: str, attrs: dict[str, str]) -> str | None:
"""Extract function name from various formats"""
if attrs and "name" in attrs:
return attrs["name"]

if "=" in name:
parts = name.split("=", 1)
if len(parts) == 2 and parts[0] == "function":
return parts[1]

return None

def _extract_parameter_name(self, name: str, attrs: dict[str, str]) -> str | None:
"""Extract parameter name from various formats"""
if attrs and "name" in attrs:
return attrs["name"]

if "=" in name:
parts = name.split("=", 1)
if len(parts) == 2 and parts[0] == "parameter":
return parts[1]

return None

def _get_param_type(self, param_name: str) -> str:
"""Get parameter type based on tool configuration, defaults to string
Args:
param_name: Parameter name
Returns:
Parameter type
"""
if not self.tools or not self.current_function_name:
return "string"

for tool in self.tools:
if not hasattr(tool, "type") or not (
hasattr(tool, "function") and hasattr(tool.function, "name")
):
continue
if (
tool.type == "function"
and tool.function.name == self.current_function_name
):
if not hasattr(tool.function, "parameters"):
return "string"
params = tool.function.parameters
if isinstance(params, dict) and "properties" in params:
properties = params["properties"]
if param_name in properties and isinstance(
properties[param_name], dict
):
return self.repair_param_type(
str(properties[param_name].get("type", "string"))
)
elif isinstance(params, dict) and param_name in params:
param_config = params[param_name]
if isinstance(param_config, dict):
return self.repair_param_type(
str(param_config.get("type", "string"))
)
break
return "string"

def _get_effective_param_type(self, param_name: str) -> str:
"""Get effective parameter type
Args:
param_name: Parameter name
Returns:
Effective parameter type
"""
# DeepSeek: if explicit type annotation exists, use it
if (
hasattr(self, "current_param_explicit_type")
and self.current_param_explicit_type
):
return self.current_param_explicit_type

# Otherwise infer from tool schema
return self._get_param_type(param_name)

def repair_param_type(self, param_type: str) -> str:
"""Repair unknown parameter types by treating them as string
Args:
param_type: Parameter type
Returns:
Repaired parameter type
"""
if (
param_type in ["string", "str", "text", "varchar", "char", "enum"]
or param_type.startswith("int")
or param_type.startswith("uint")
or param_type.startswith("long")
or param_type.startswith("short")
or param_type.startswith("unsigned")
or param_type.startswith("num")
or param_type.startswith("float")
or param_type in ["boolean", "bool", "binary"]
or (
param_type in ["object", "array", "arr", "sequence"]
or param_type.startswith("dict")
or param_type.startswith("list")
)
):
return param_type
else:
return "string"

def _convert_param_value(self, param_value: str, param_type: str) -> Any:
"""Convert value based on parameter type
Args:
param_value: Parameter value
param_type: Parameter type
Returns:
Converted value
"""
if param_value.lower() == "null":
return None

param_type = param_type.strip().lower()
if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
return param_value
elif (
param_type.startswith("int")
or param_type.startswith("uint")
or param_type.startswith("long")
or param_type.startswith("short")
or param_type.startswith("unsigned")
):
try:
return int(param_value)
except (ValueError, TypeError):
logger.warning(
"Parsed value '%s' of parameter '%s' is not an integer "
"in tool '%s', degenerating to string.",
param_value,
)
return param_value
elif param_type.startswith("num") or param_type.startswith("float"):
try:
float_param_value: float = float(param_value)
return (
float_param_value
if float_param_value - int(float_param_value) != 0
else int(float_param_value)
)
except (ValueError, TypeError):
logger.warning(
"Parsed value '%s' of parameter '%s' is not a float "
"in tool '%s', degenerating to string.",
param_value,
)
return param_value
elif param_type in ["boolean", "bool", "binary"]:
param_value = param_value.lower()
return param_value == "true"
else:
return param_value

def _convert_for_json_streaming(self, converted_value: Any, param_type: str) -> str:
"""Convert converted_value based on
whether it's empty and if type is string
Args:
converted_value: Converted value
param_type: Parameter type
Returns:
Converted string for streaming output
"""
# Check if value is empty, but exclude numeric 0
if converted_value is None or converted_value == "":
return ""

if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
# String type, remove double quotes
return json.dumps(converted_value, ensure_ascii=False)[1:-1]
else:
# Non-string type, return complete JSON string
if not isinstance(converted_value, str):
return json.dumps(converted_value, ensure_ascii=False)
else:
return converted_value

def _reset_xml_parser_after_tool_call(self):
"""
Each tool_call is treated as a separate XML document,
so we need to reset the parser after each tool_call.
"""

# recreate XML parser
self.parser = ParserCreate()
self.setup_parser()

# Reset current tool_call state
if self.current_call_id:
self.last_completed_call_id = self.current_call_id
self.current_call_id = None
self.current_function_name = None
self.current_function_open = False
self.parameters = {}
self.current_param_name = None
self.current_param_value = ""
self.current_param_value_converted = ""
self.current_param_is_first = False
self.should_emit_end_newline = False
self.start_quote_emitted = False
self.text_content_buffer = ""

# Reset preprocessing and deferred parsing state
self._pre_inside_parameter = False
self._pre_param_buffer = ""
self._pre_current_param_name = None
self.defer_current_parameter = False
self.deferred_param_raw_value = ""

def _reset_current_call_state(self):
"""
Lightweight reset for current invoke state.
Used for DeepSeek to handle multiple invokes within the same
function_calls container.
Does NOT reset the parser or deltas.
"""
# Reset current tool_call state
if self.current_call_id:
self.last_completed_call_id = self.current_call_id
self.current_call_id = None
self.current_function_name = None
self.current_function_open = False
self.parameters = {}
self.current_param_name = None
self.current_param_value = ""
self.current_param_value_converted = ""
self.current_param_is_first = False
self.should_emit_end_newline = False
self.start_quote_emitted = False
self.text_content_buffer = ""

# Reset preprocessing and deferred parsing state
self._pre_inside_parameter = False
self._pre_param_buffer = ""
self._pre_current_param_name = None
self.defer_current_parameter = False
self.deferred_param_raw_value = ""

# Reset explicit type
self.current_param_explicit_type = None

# Note: Do NOT reset:
# - self.parser (XML parser instance)
# - self.deltas (accumulated delta list)
# - self.tool_call_index (tool call index continues to increment)
# - self.streaming_buffer (streaming buffer)

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

The StreamingXMLToolCallParser class is extremely complex. It implements a stateful, streaming XML parser with deferred parsing logic, fallback mechanisms, and manual buffer management, essentially building a parser on top of Python's expat parser. This complexity makes the code very difficult to understand, debug, and maintain. For instance, the interaction between _find_next_complete_element, _process_complete_xml_elements, and _preprocess_xml_chunk creates a convoluted data flow that is hard to trace.

While parsing model-generated tool calls in a streaming fashion is inherently complex, this implementation seems overly so. It's worth investigating if a simpler design is possible that relies more on the expat parser's built-in streaming capabilities and reduces the amount of manual state management and pre-parsing logic. A simpler, more declarative implementation would be more robust and easier to maintain in the long run. Since this code is reused from another parser, consider refactoring the common logic into a shared base class to improve maintainability and avoid code duplication.

Comment on lines 1 to 184
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project

import json

import pytest

from vllm.entrypoints.openai.protocol import (
ChatCompletionRequest,
ChatCompletionToolsParam,
FunctionCall,
ToolCall,
)
from vllm.entrypoints.openai.tool_parsers.deepseek_v32_tool_parser import (
DeepSeekV32ToolParser,
)
from vllm.tokenizers import get_tokenizer

pytestmark = pytest.mark.cpu_test

MODEL = "deepseek-ai/DeepSeek-V3"


@pytest.fixture(scope="module")
def deepseek_tokenizer():
return get_tokenizer(tokenizer_name=MODEL)


@pytest.fixture
def deepseek_tool_parser(deepseek_tokenizer):
return DeepSeekV32ToolParser(deepseek_tokenizer)


@pytest.fixture
def sample_tools():
return [
ChatCompletionToolsParam(
type="function",
function={
"name": "get_weather",
"description": "Get the weather for a location",
"parameters": {
"type": "object",
"properties": {
"location": {"type": "string", "description": "The city name"},
"date": {"type": "string", "description": "The date"},
},
"required": ["location", "date"],
},
},
),
]


def assert_tool_calls(
actual_tool_calls: list[ToolCall], expected_tool_calls: list[ToolCall]
):
assert len(actual_tool_calls) == len(expected_tool_calls)

for actual_tool_call, expected_tool_call in zip(
actual_tool_calls, expected_tool_calls
):
assert actual_tool_call.type == "function"
assert actual_tool_call.function.name == expected_tool_call.function.name
try:
assert json.loads(actual_tool_call.function.arguments) == json.loads(
expected_tool_call.function.arguments
)
except json.JSONDecodeError as e:
print(e)
print("actual_tool_call", actual_tool_call.function.arguments)
print("expected_tool_call", expected_tool_call.function.arguments)


def test_extract_tool_calls_single_function(
deepseek_tool_parser,
sample_tools,
):
"""Test extracting a single function call"""
model_output = """<|DSML|function_calls>
<|DSML|invoke name="get_weather">
<|DSML|parameter name="location" string="true">Hangzhou</|DSML|parameter>
<|DSML|parameter name="date" string="true">2024-01-16</|DSML|parameter>
</|DSML|invoke>
</|DSML|function_calls>"""

expected_tool_calls = [
ToolCall(
function=FunctionCall(
name="get_weather",
arguments=json.dumps({"location": "Hangzhou", "date": "2024-01-16"}),
)
),
]

request = ChatCompletionRequest(
model="deepseek-v3",
messages=[{"role": "user", "content": "What is the weather?"}],
tools=sample_tools,
)

extracted = deepseek_tool_parser.extract_tool_calls(model_output, request)

assert extracted.tools_called
assert extracted.content is None
assert_tool_calls(extracted.tool_calls, expected_tool_calls)


def test_extract_tool_calls_multiple_functions(
deepseek_tool_parser,
sample_tools,
):
"""Test extracting multiple function calls"""
model_output = """<|DSML|function_calls>
<|DSML|invoke name="get_weather">
<|DSML|parameter name="location" string="true">Hangzhou</|DSML|parameter>
<|DSML|parameter name="date" string="true">2024-01-16</|DSML|parameter>
</|DSML|invoke>
<|DSML|invoke name="get_weather">
<|DSML|parameter name="location" string="true">Beijing</|DSML|parameter>
<|DSML|parameter name="date" string="true">2024-01-16</|DSML|parameter>
</|DSML|invoke>
</|DSML|function_calls>"""

expected_tool_calls = [
ToolCall(
function=FunctionCall(
name="get_weather",
arguments=json.dumps({"location": "Hangzhou", "date": "2024-01-16"}),
)
),
ToolCall(
function=FunctionCall(
name="get_weather",
arguments=json.dumps({"location": "Beijing", "date": "2024-01-16"}),
)
),
]

request = ChatCompletionRequest(
model="deepseek-v3",
messages=[{"role": "user", "content": "What is the weather?"}],
tools=sample_tools,
)

extracted = deepseek_tool_parser.extract_tool_calls(model_output, request)

assert extracted.tools_called
assert extracted.content is None
assert_tool_calls(extracted.tool_calls, expected_tool_calls)


def test_extract_tool_calls_with_end_of_sentence_token(
deepseek_tool_parser,
sample_tools,
):
"""Test extracting function calls with end-of-sentence token"""
model_output = """<|DSML|function_calls>
<|DSML|invoke name="get_weather">
<|DSML|parameter name="location" string="true">Hangzhou</|DSML|parameter>
<|DSML|parameter name="date" string="true">2024-01-16</|DSML|parameter>
</|DSML|invoke>
</|DSML|function_calls><|end▁of▁sentence|>"""

expected_tool_calls = [
ToolCall(
function=FunctionCall(
name="get_weather",
arguments=json.dumps({"location": "Hangzhou", "date": "2024-01-16"}),
)
),
]

request = ChatCompletionRequest(
model="deepseek-v3",
messages=[{"role": "user", "content": "What is the weather?"}],
tools=sample_tools,
)

extracted = deepseek_tool_parser.extract_tool_calls(model_output, request)

assert extracted.tools_called
assert extracted.content is None
assert_tool_calls(extracted.tool_calls, expected_tool_calls)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

The added tests cover the basic cases of single and multiple function calls. However, they do not cover more complex edge cases that the parser is designed to handle, or which might expose its fragility. Given the complexity of the XML streaming parser, it's important to have comprehensive tests for:

  • Parameter values containing special XML characters (e.g., <, >, &, ', ").
  • Parameter values that are complex objects or arrays, especially those that would trigger the deferred parsing logic (e.g., containing Python literals with single quotes).
  • Malformed XML output from the model (e.g., unclosed tags).
  • Text content mixed with tool calls.

Adding these tests would significantly improve the robustness and maintainability of the tool parser.

@github-actions
Copy link

github-actions bot commented Dec 2, 2025

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors.

You ask your reviewers to trigger select CI tests on top of fastcheck CI.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

🚀

Signed-off-by: wenmengzhou <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

deepseek Related to DeepSeek models frontend qwen Related to Qwen models tool-calling

Projects

Status: No status

Development

Successfully merging this pull request may close these issues.

1 participant