Uncensored Multi-Model Releases, LongCat-Flash-Lite-Sparse with MTPs and LSAs, Qwen3.8-27B with MTPs, Qwen3.5-122B-A10B with MTPs, Qwen3-Coder-Next and Laguna-S2.1 with Vision, All Available in GGU...

Been working really hard for the past month to bring to the community all these models, the hardest was for sure LongCat-Flash-Lite-Sparse who required TONS of work, first I needed to have Heretic support created for it from scratch and had to create support for it on llama.cpp too, quite difficult and time consuming task! It was even more difficult to work on than the original LongCat-Flash-Lite model that I released a few weeks ago, it is still a 69B-A3B model as the original LongCat-Flash-Lite, but LongCat-Flash-Lite-Sparse has now added support for: - Sparse attention (vs dense attention for LongCat-Flash-Lite) - 1M Context length (vs 256k for LongCat-Flash-Lite) Anyway LongCat-Flash-Lite-Sparse has 0 support on mainline/upstream llama.cpp, so to be able to use the GGUFs you will need to pull my fork from GitHub, which you can find here: github.com/erm14254/llama.cpp-minimax-m3-combined/tree/claude/longcat-win11 You would need to load the model through llama-server.exe and you can interact with it through llama-ui. You have two variants, Uncensored Heretic (9/100 refusals for 0.0157 KLD) and Ultra Uncensored HJeretic (4/100 refusals for 0.0779 KLD), both variants come with MTPs and LSAs! Here is the model links: Uncensored Heretic GGUFs: huggingface.co/llmfan46/LongCat-Flash-L...etic-Native-MTP-And-LSA-Preserved-GGUF Ultra Uncensored Heretic GGUFs: huggingface.co/llmfan46/LongCat-Flash-L...etic-Native-MTP-And-LSA-Preserved-GGUF ---------------------------------------- That's it for LongCat, so next we have Qwen3.8-27B Ultra Uncensored Heretic with MTPs , 3/100 refusals for 0.0244 KLD, you can find the links here: Safetensors: huggingface.co/llmfan46/Qwen3.8-27B-Ult...ncensored-Heretic-Native-MTP-Preserved GGUFs: huggingface.co/llmfan46/Qwen3.8-27B-Ult...ored-Heretic-Native-MTP-Preserved-GGUF NVFP4: huggingface.co/llmfan46/Qwen3.8-27B-Ult...red-Heretic-Native-MTP-Preserved-NVFP4 NVFP4 GGUFs: huggingface.co/llmfan46/Qwen3.8-27B-Ult...eretic-Native-MTP-Preserved-NVFP4-GGUF GPTQ-Int4: huggingface.co/llmfan46/Qwen3.8-27B-Ult...Heretic-Native-MTP-Preserved-GPTQ-Int4 ---------------------------------------- Next we have Qwen3.5-122B-A10B Uncensored Heretic with MTPs , 8/100 refusals for 0.0856 KLD, here: GGUFs: huggingface.co/llmfan46/Qwen3.5-122B-A1...ored-Heretic-Native-MTP-Preserved-GGUF ---------------------------------------- Then we have Qwen3-Coder-Next , which is a model that was requested by a Hugging Face user some time ago, so I finally had time to work on it, here is the link: GGUFs: huggingface.co/llmfan46/Qwen3-Coder-Next-Uncensored-Heretic-GGUF ---------------------------------------- And finally Laguna-S2.1 with Vision , get it from here: GGUFs: huggingface.co/llmfan46/Laguna-S-2.1-Uncensored-Heretic-Vision-GGUF The visions part is far from perfect, so if you do not want to use vision you can simply not download the mmproj files and the model will just function like a regular text-only model. ---------------------------------------- I also made some improvements to J-Wash by adding support for MoE Qwen3.5/3.6/3.8 models support, improvments, bug fixes, improvements to the UI to make it easier to use and more practical for users etc, in case you are interested here is the link: github.com/erm14254/J-Wash-Enhanced/tree/master ---------------------------------------- That's it for now! As usual you can find all my models here: HuggingFace-LLMFan46 Tremendous amount of work went into making these releases come true, so if you like my work and find my models useful, then I would really appreciate if you could support me on Ko-fi: ko-fi.com/llmfan46