GLM-4.6 GGUF (Unsloth Dynamic 2.0)
This is a GGUF conversion of Zhipu’s GLM-4.6 packaged by Unsloth, intended for running locally with llama.cpp-style tooling (or apps built on top of it). The model card emphasizes two practical details: (1) you’ll want a recent llama.cpp, and (2) the GGUF includes multiple chat-template fixes, including support for Jinja templates.
From a capability standpoint, GLM-4.6 is positioned as an incremental step up from GLM-4.5 with a longer context window (200K tokens), improved coding performance, and stronger reasoning/tool-use behavior. That combination is especially relevant if you’re evaluating “agentic” coding assistants, because a bigger working context plus better tool use can materially change how often a model gets lost in multi-step tasks.
If you want to try it quickly: pick a quant that fits your CPU/RAM (or GPU VRAM if you’re using a GPU-accelerated runtime), then run a small coding benchmark prompt you already trust (bugfix, refactor, or “add tests” request). Keep temperature and sampling parameters consistent so you can compare runs across models.
Quick stats from the listing feed: pipeline: text-generation · 144 likes · 35388 downloads.
Source listing: https://huggingface.co/models?sort=modified