Blackfrost-Research/KIMI-K3-DERISKED-MXFP4-GGUF
Text Generation • 2.8T • Updated • 1
2.8T LatentMoE + KDA, refusal-surface reduced at the weight level (ABLITERATED). GGUF variants need llama.cpp PR #26185 — mainline cannot load them.
Note Byte-neutral GGUF repack, experts bit-exact
Note Free tier - all-Q2_K, fits one 8xB200 node, ~940 GiB