From 7fd4acc35ca513f5e62dbfdd460309337e7052e6 Mon Sep 17 00:00:00 2001 From: Concedo <39025047+LostRuins@users.noreply.github.com> Date: Tue, 11 Aug 2026 15:43:23 +0800 Subject: [PATCH] reasoning budget for muse glimmer --- gpttype_adapter.cpp | 6 ++++++ koboldcpp.py | 4 ++++ 2 files changed, 10 insertions(+) diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp index af9397583..8729bfbdf 100644 --- a/gpttype_adapter.cpp +++ b/gpttype_adapter.cpp @@ -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; } diff --git a/koboldcpp.py b/koboldcpp.py index b64d89a21..d5d5b84b6 100644 --- a/koboldcpp.py +++ b/koboldcpp.py @@ -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