Marx

I publish models, adapters, and research notes from my local ML experiments. Mostly LLM internals — abliteration, disinhibition direction geometry, LoRA fine-tuning, and inference optimization. Most of what's here falls into a few buckets: - **Model edits** — abliterated and disinhibited variants, direction-based behavioral editing, row-norm restoration experiments - **LoRA adapters** — fine-tuned on curated reasoning traces, geometric abliteration adapters - **Inference research** — IREE VMFB compilation pipelines for vision and language models Everything is experimental. Models are published when they're interesting, not when they're polished. No support, no roadmap, no guarantees. If someone can make use of any of my findings or models feel free to use them however you want (or the licensing allows) Self-taught. Based in Germany. Works in retail.