Tag
GLM-5.3 and GLM-5.3-Flash received updated HuggingFace chat templates that optimize tool-result reordering with an early exit, and developers are advised to pull the latest template for deployment.
The author experimented with the 27B model on VLLM and created a 'high' reasoning mode by blending prompts from low and xhigh modes, resulting in more efficient and enjoyable reasoning output.
This article details the improvements to the Jinja chat template for Qwen 3.8 models, correcting issues from previous versions to ensure consistent output quality and benchmark performance.
A PSA about Meta's Muse Glimmer: setting reasoning strength via system prompt is partially overridden by the chat template's high default, and the absent default is high, not off. Using chat_template_kwargs.reasoning_strength provides more control.
A PSA warning DeepSeek-V4-Flash-0731 users that system role messages are hoisted to the top, ruining prompt caching and proximity, and advising the use of latest_reminder instead.
Recent llama.cpp commits broke preserve_thinking behavior for older DeepSeek V4 gguf chat templates, causing issues in coding agent contexts. The fix is to override the gguf template with a new one using --chat-template-file.
Laguna S-2.1 model has been updated with a fix for yarn_attn_factor (corrected to 1.0) and an improved chat template that fixes broken thinking, preserves thinking, and enables tool calling. Users are advised to use the updated GGUF from the official repo.
Reports a bug in Laguna S 2.1 where reasoning phase fails if preserve_thinking is disabled, and suggests comparing with Qwen 27B chat template to fix the issue.
An analysis shows that Gemma 4 models' benchmark performance is heavily influenced by chat templates rather than model weights, with template changes causing behavioral shifts without altering any parameters; notably, all sizes fail a crisis-signal scenario.
The user shares plans to use Sol for optimizing MTP support in llama.cpp, standardizing chat templates, reviving an abandoned whisper project, finishing a personal finance app, and kernel optimization.
Jackrong releases Qwopus3.6-27B-Coder-Compat-MTP-GGUF, a GGUF quantization of the Qwopus3.6-27B-Coder model with an expanded chat template for better interoperability with tool-using runtimes and OpenAI-compatible agent frameworks.
Google's Gemma 4 31B IT model now has a chat template fix that preserves thinking and improves null handling, reasoning preservation, and input validation.
Proposes the Piggyback Hypothesis that chat-template tokens can cause emergent misalignment in LLMs, and introduces Token-Regularized Finetuning (TReFT) to mitigate it while preserving in-domain learning.
Gemma 4 12B has a known issue with tool calling and coding, but using a custom chat template in llama.cpp resolves the bugs. Users should compile llama.cpp from source and apply the fix before evaluating the model's coding ability.
A new web tool, Chat Template Playground, lets users visualize how different open-source LLMs render their chat templates, highlighting differences in prompting and tokenization.
This paper investigates why instruction-tuned LLMs are overconfident in their own responses, identifying an 'ownership bias' that gives higher confidence to self-generated answers. It proposes a simple inference-time strategy to reframe the model's answer as user input, improving calibration by up to 26% without retraining.
This paper systematically investigates when and which parts of the prompt KV cache become dispensable during LLM decoding, showing that redundancy primarily involves chat template scaffolding rather than task content, and replacement with neutral filler preserves accuracy.
A developer shares that addressing the 'tito' problem in multi-turn reinforcement learning is simpler than commonly believed, requiring only one implementation rule and a chat-template property that models already support.
The author tests Qwen 3.6 27b F16 on a Pacman coding benchmark, achieving near-perfect results in two of three attempts, while noting the critical impact of quantization, chat templates, and speculative decoding for local agentic coding.
This repository provides fixed Jinja chat templates for Qwen 3.5 and 3.6, addressing rendering errors, token waste, and missing features in the official templates for engines like LM Studio and llama.cpp.