[Bugfix][Frontend] Respect max_output_tokens in the Harmony tool-call loop (#58551)

Signed-off-by: errmakov <ide404@gmail.com>
Co-authored-by: Du Bin <8174807+dubin555@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Flora Feng <4florafeng@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Denis Ermakov
2026-09-25 14:00:39 -07:00
committed by GitHub
co-authored by Du Bin Claude Opus 5.5 Flora Feng mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 83db8c8390
commit 7871963fcc
3 changed files with 57 additions and 2 deletions
@@ -615,12 +615,14 @@ class HarmonyContext(ConversationContext):
available_tools: list[str],
function_tool_names: frozenset[str],
response_parser: Parser | None = None,
request: ResponsesRequest | None = None,
):
from vllm.parser.harmony import HarmonyParser, Segment
assert isinstance(response_parser, HarmonyParser)
self._messages = messages
self.request = request
self.response_parser: HarmonyParser = response_parser
self.finish_reason: str | None = None
self.available_tools = available_tools
+7 -2
View File
@@ -443,6 +443,7 @@ class OpenAIServingResponses(GenerateBaseServing):
available_tools,
function_tool_names,
response_parser=response_parser,
request=request,
)
else:
if envs.VLLM_USE_EXPERIMENTAL_PARSER_CONTEXT:
@@ -678,8 +679,12 @@ class OpenAIServingResponses(GenerateBaseServing):
tok_params=tok_params,
)
sampling_params.max_tokens = max_model_len - self._extract_prompt_len(
engine_input
sampling_params.max_tokens = get_max_tokens(
max_model_len,
context.request.max_output_tokens if context.request else None,
self._extract_prompt_len(engine_input),
self.default_sampling_params,
self.override_max_tokens,
)
elif isinstance(context, ParsableContext):
(engine_input,) = await self._render_next_turn(