Downloads · 30 days
76
9% of all-time downloads
robinshao/hauhau_qwen36_35b_a3b_mxfp4_gguf_mtp
hauhau_qwen36_35b_a3b_mxfp4_gguf_mtp is a text generation model from robinshao. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
This repo contains a local MXFP4MOE GGUF export of HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive together with a separate MTP draft GGUF for speculative decoding in a patched local llama.cpp toolchain.
Downloads · 30 days
76
9% of all-time downloads
All-time downloads
812
Public
Repo size
20.6 GB
Likes
0
Public
Click a slice to open those files.
.gguf20.6 GB · 100%
From the Hugging Face model README
This repo contains a local MXFP4_MOE GGUF export of HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive together with a separate MTP draft GGUF for speculative decoding in a patched local llama.cpp toolchain.
hauhau_qwen36_35b_a3b_mxfp4_robinstyle.gguf: main text model in robinshao-style MXFP4 layouthauhau_qwen36_35b_a3b_mxfp4_mtp.gguf: MTP draft GGUF used with --spec-type draft-mtpllama.cpp.src-patched.zip: patched local llama.cpp source snapshot used for conversion and servingbuild_result.json: build and validation summary for the main model and MTP draftmtp-benchmark.json: short local speed comparison between baseline and MTP servingkhoi11/Qwen3.6-35B-A3B-Uncensored-HauhauCS-FP16HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-AggressiveMXFP4_MOE GGUFllama.cppThis pair was validated locally on Windows with patched llama.cpp using speculative decoding:
llama-server -m hauhau_qwen36_35b_a3b_mxfp4_robinstyle.gguf \
--spec-type draft-mtp \
--spec-draft-model hauhau_qwen36_35b_a3b_mxfp4_mtp.gguf \
--spec-draft-device CUDA0 \
--spec-draft-ngl all \
--jinja \
--device CUDA0 \
--gpu-layers all \
--ctx-size 8192 \
--batch-size 768 \
--ubatch-size 128 \
--flash-attn on \
--cache-type-k f16 \
--cache-type-v f16 \
--cache-ram 0
In a short local API benchmark on an RTX 4000 Ada, the MTP setup improved decode speed from about 103.45 tok/s to about 131.65 tok/s, roughly 1.27x.
262144, but the actual no-offload local limit depends on GPU VRAM, KV cache settings, and whether the draft model is loaded.