AI, automation & information security
Patrick Gawron
Agentic workflows, local-first AI, secure business logic. Daily AI experiments, consulting projects - and a build log of coding sidequests.
Tools
Notes
Quick posts for everything that doesn't need a full article - local LLMs, CSS, keto. One page each.
Browse all notes >Tech news
Bartowski reworks GGUF tensor layouts for sharper quantised models
New per-tensor layout maps change how weights are packed in GGUF quants, per bartowski's tests improving quality at the same bit-width - relevant if you pull his quants for llama.cpp.
DeepSeek ships V4.1 Flash, a 552B multimodal MoE model under MIT
552B-parameter multimodal MoE under MIT, context up to 1M tokens. Full weights don't fit one consumer card, but the permissive licence means community GGUF quantisations can follow fast.
YuE2 ships a 3B open-weights music model with symbolic planning
3B parameters, runs on a single consumer GPU, plans song structure symbolically before generating audio - a full text-to-music stack you can self-host instead of renting Suno.
Lightricks opens LTX 2.5: 22B video weights that make their own audio
Open weights, free commercially under $10M ARR, ComfyUI support on day zero, and video and sound come out of one diffusion pass. On a single 32 GB card the text encoder and the transformer do not fit together, so every changed word in the prompt costs 27 seconds before the first frame.