From 1087676a905d283aa71731fd6b94e10e91f58add Mon Sep 17 00:00:00 2001 From: Flora Feng <4florafeng@gmail.com> Date: Wed, 13 May 2026 20:35:45 -0400 Subject: [PATCH] [Refactor] Use shared utils in hermes tool parser (#42570) Signed-off-by: sfeng33 <4florafeng@gmail.com> --- vllm/tool_parsers/hermes_tool_parser.py | 45 ++++++------------------- 1 file changed, 11 insertions(+), 34 deletions(-) diff --git a/vllm/tool_parsers/hermes_tool_parser.py b/vllm/tool_parsers/hermes_tool_parser.py index 469bc8363b..546cde5cd1 100644 --- a/vllm/tool_parsers/hermes_tool_parser.py +++ b/vllm/tool_parsers/hermes_tool_parser.py @@ -25,33 +25,20 @@ from vllm.tool_parsers.abstract_tool_parser import ( Tool, ToolParser, ) +from vllm.tool_parsers.utils import is_complete_json, partial_tag_overlap from vllm.utils.mistral import is_mistral_tokenizer logger = init_logger(__name__) -def _partial_tag_overlap(text: str, tag: str) -> int: - """Length of the longest prefix of `tag` that matches a suffix of `text`. - - E.g. text ending in "". - Returns 0 if there is no overlap. - """ - max_check = min(len(tag) - 1, len(text)) - for k in range(max_check, 0, -1): - if text.endswith(tag[:k]): - return k - return 0 - - -def _is_valid_json(text: str) -> bool: - try: - json.loads(text) - return True - except (json.JSONDecodeError, ValueError): - return False - - class Hermes2ProToolParser(ToolParser): + tool_call_start_token: str = "" + tool_call_end_token: str = "" + tool_call_regex = re.compile( + r"(.*?)|(.*)", re.DOTALL + ) + scratch_pad_regex = re.compile(r"(.*?)", re.DOTALL) + def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None): super().__init__(tokenizer, tools) @@ -59,16 +46,6 @@ class Hermes2ProToolParser(ToolParser): logger.error("Detected Mistral tokenizer when using a Hermes model") self.model_tokenizer = tokenizer.tokenizer - self.tool_call_start_token: str = "" - self.tool_call_end_token: str = "" - - self.tool_call_regex = re.compile( - r"(.*?)|(.*)", re.DOTALL - ) - self.scratch_pad_regex = re.compile( - r"(.*?)", re.DOTALL - ) - if not self.model_tokenizer: raise ValueError( "The model tokenizer must be passed to the ToolParser " @@ -147,7 +124,7 @@ class Hermes2ProToolParser(ToolParser): Holds back any suffix that could be a partial tag. """ if self.tool_call_start_token not in current_text: - overlap_length = _partial_tag_overlap( + overlap_length = partial_tag_overlap( current_text, self.tool_call_start_token ) sendable_idx = len(current_text) - overlap_length @@ -176,13 +153,13 @@ class Hermes2ProToolParser(ToolParser): else: raw = text[json_start:] # Strip partial suffix if present. - overlap = _partial_tag_overlap(raw, self.tool_call_end_token) + overlap = partial_tag_overlap(raw, self.tool_call_end_token) if overlap: raw = raw[:-overlap] tc_json = raw.strip() # Valid JSON without closing tag = complete body, # tag tokens just haven't arrived yet. - is_complete = _is_valid_json(tc_json) if tc_json else False + is_complete = is_complete_json(tc_json) if tc_json else False results.append((tc_json, is_complete)) break return results