Bathory X's experimental writing model gets local GGUF builds
An October 12 community merge aimed at prose and roleplay now has downloadable imatrix GGUF variants. The creator's writing-quality claims remain personal tests.
A community merge reaches GGUF
Bathory X 26B-A4B v0.2, an experimental model for prose and roleplay, now has weighted imatrix GGUF builds for people running models locally. The October 13 repository from mradermacher contains downloadable variants spanning low-bit IQ formats through Q6, plus the imatrix file used to make them. This is a quantization of an existing October 12 model release, not a new foundation model or an official update from any of its component developers. The original checkpoint remains in Nabbers1999’s repository.
What the creator claims
The original English model card describes a merge of body layers from several community models, output-head distillation on prose and roleplay data, and edits meant to reduce refusals. Its author calls v0.2 a substantial writing improvement over v0.1, citing personal testing rather than a published benchmark. The GGUF card documents file choices and sizes but does not establish how each format changes writing quality. AITrending has checked the public cards and file listings, not run the models or verified the creator’s quality comparison. Users choosing a build should also check the upstream model’s terms and their own hardware limits.