reasoning budget for muse glimmer

This commit is contained in:
Concedo 2026-08-11 15:43:23 +08:00
parent dcfe29317e
commit 7fd4acc35c
2 changed files with 10 additions and 0 deletions

View file

@ -5881,6 +5881,12 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
end = "[/THINK]";
budget_exceeded = "\n(Reasoning budget exceeded)\nTime to respond now.\n[/THINK]";
break;
case llm_arch::LLM_ARCH_MUSE_GLIMMER:
start = " to=self<|message|>";
end = "<|eom|>";
budget_exceeded = "\n(Reasoning budget exceeded)\nTime to respond now.\n<|eom|>";
expected_start_tokens = 3;
break;
default:
break;
}

View file

@ -4623,6 +4623,10 @@ ws ::= | " " | "\n" [ \t]{0,20}
copied_jinja_kwargs.update(user_chat_template_kwargs)
if "reasoning_effort" in genparams and genparams["reasoning_effort"] is not None:
copied_jinja_kwargs["reasoning_effort"] = genparams["reasoning_effort"]
if "enable_thinking" in copied_jinja_kwargs and not copied_jinja_kwargs["enable_thinking"]:
copied_jinja_kwargs["reasoning_strength"] = "none"
elif "reasoning_effort" in copied_jinja_kwargs and copied_jinja_kwargs["reasoning_effort"]:
copied_jinja_kwargs["reasoning_strength"] = copied_jinja_kwargs["reasoning_effort"]
jinja_output = format_jinja(messages_array,jinjatools,copied_jinja_kwargs)
if jinja_output:
messages_string = jinja_output