base_model:
- edwixx/diffusiongemma-26B-A4B-it-HERETIC-Uncensored
pipeline_tag: text-generation
tags:
- diffusion
- fp8
- text-generation-inference
DiffusionGemma-26B HERETIC (FP8-Dynamic)
26.6B params · 4B active · MoE
This is an FP8 quantized and deduplicated version of edwixx/diffusiongemma-26B-A4B-it-HERETIC-Uncensored, quantized using the llm-compressor library. The checkpoint size is ~26.1 GB (single shard model.safetensors), down from the original ~52 GB
Legal Disclaimer & Terms of Use (click to expand)
THIS MODEL IS PROVIDED FOR RESEARCH AND EDUCATIONAL PURPOSES ONLY. I am not a lawyer; this disclaimer is not legal advice; the User should seek their own counsel.
Access to these weights requires acceptance of these terms via HuggingFace gated access. By requesting and being granted access, downloading, using, distributing, hosting, modifying, redistributing, sublicensing, or otherwise accessing this model, or any output generated by this model, you (the "User") acknowledge and agree to the following:
Intended Use and Not a Service
This model is a static release of model weights for research, safety evaluation, interpretability, and red-teaming. I distribute model weights only and do not host, operate, or provide inference access to this model. I am not a "deployer" or "operator" of any AI system. No hosted endpoint, API, demo, Space, or interactive interface is provided or endorsed by me.
No Intent to Facilitate Harm
I do not intend, encourage, or design this model to be used in the commission of any crime. The removal of refusal behavior is a research result regarding model interpretability and alignment, not an endorsement or facilitation of any unlawful act.