Qwen3.8-27B-MTP-TQ3_4S
TurboQuant TQ3_4S GGUF build of Qwen/Qwen3.8-27B with native MTP (Multi-Token Prediction) draft heads for speculative decoding.
This file was quantized with the TurboQuant toolchain and is intended for the public TurboQuant runtime fork maintained by Y Tan (GitHub: turbo-tan):
- Runtime fork: https://github.com/turbo-tan/llama.cpp-tq3
- Reference releases by Y Tan: https://huggingface.co/YTan2000
Credit goes to Y Tan for the TQ3_4S quantization path and the llama.cpp runtime support this model depends on. See the Qwen3.5-27B-TQ3_4S reference card for the format's published quality and speed characteristics.
Required Runtime
This model requires the TurboQuant runtime fork — stock llama.cpp builds cannot load the TQ3_4S tensor type.
This is an MTP model: the GGUF includes the trained MTP draft block, so serve it with draft-MTP speculative decoding for full decode speed.
git clone https://github.com/turbo-tan/llama.cpp-tq3.git
cd llama.cpp-tq3
cmake -B build -DGGML_CUDA=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build -j
File
- File:
Qwen3.8-27B-MTP-TQ3_4S.gguf - Size: 13.68 GB (12.74 GiB)
- SHA256:
b734f643665a9d1f356cfcca0ef65572eb08ab9ab25c29b1ddf4ac602548e2cb - Quantization:
TQ3_4S(3.5-bit Walsh-Hadamard transform format) - Architecture:
qwen35— 64 layers, hybrid Gated DeltaNet + Gated Attention, 1 MTP draft block - Context: 262,144 tokens native
Recommended Launch
./build/bin/llama-server \
-m Qwen3.8-27B-MTP-TQ3_4S.gguf \
--host 127.0.0.1 --port 8080 \
-c 32768 -np 1 -ngl 99 -fa on \
-ctk q8_0 -ctv tq3_0 \
--spec-type draft-mtp \
--spec-draft-n-min 1 \
--spec-draft-n-max 2 \
--spec-draft-p-min 0.0 \
--no-spec-draft-backend-sampling \
--reasoning off --jinja
Settings notes:
--spec-draft-n-max 2is optimal for the 3.8 MTP head (measured)--no-spec-draft-backend-samplingkeeps CUDA graph reuse high- The model supports thinking mode via
--reasoning-format deepseek --reasoning-budget N
License
Subject to the Qwen3.8-27B base model license (Apache 2.0) and the license terms of the runtime components used to run the GGUF.