[Frontend] Parse tool calls and reasoning from checkpoint response templates (#58604)

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Yoni Gozlan <yonigozlan@users.noreply.github.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
This commit is contained in:
Yoni Gozlan
2026-10-02 02:33:00 +00:00
committed by GitHub
co-authored by Cursor Bugen Zhao Claude Opus 5.5
parent 5a661487e2
commit 9bd993564f
18 changed files with 2298 additions and 21 deletions
@@ -165,6 +165,7 @@ class OpenAIServingChat(GenerateBaseServing):
tool_strict_level=tool_strict_level,
model_name=self.model_config.model,
is_harmony=self.model_config.hf_config.model_type == "gpt_oss",
tokenizer=self.renderer.tokenizer,
)
self.exclude_tools_when_tool_choice_none = exclude_tools_when_tool_choice_none
@@ -974,6 +975,8 @@ class OpenAIServingChat(GenerateBaseServing):
parser = self._make_parser(
request, tokenizer, self._effective_chat_template_kwargs(request)
)
if parser is not None and final_res.prompt_token_ids is not None:
parser.set_prompt_token_ids(final_res.prompt_token_ids)
token_ids = output.token_ids
out_logprobs = output.logprobs