brandonbondig/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q4_K_P-FastMTP-GGUF

Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q4KP-FastMTP-GGUF

Slim mirror of HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF, published for RunPod's cached-models feature.

It carries only the two files a llama-server deployment needs to serve the Q4KP quant with FastMTP speculative decoding:

RunPod's cached model pre-downloads every file of the repo it is pointed at onto the hosts it schedules workers on. The upstream repo is 172.5 GB across ten quants, so mirroring just these two turns 172.5 GB of host-local storage into ~32 GB.

Nothing here is re-quantised or otherwise modified. The weights, the FastMTP draft and the quantisation work are HauhauCS's -- see HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF for the model card, the licence, the prompt format and the llama.cpp patch. SHA256SUMS carries the upstream checksums of these two files, so they verify against the upstream repo byte for byte.