🆕 MTP (Multi-Token Prediction) Support — All models now include the MTP / NextN head (blk.40.* tensors), enabling speculative decoding in compatible runtimes such as recent builds of llama.cpp. Each GGUF has been validated to contain the full set of MTP tensors.

🆕 Improved Quantization — All quantizations now use AutoRound iterative calibration with significantly more iterations than before, resulting in better quality across all schemes.

Quantization Details

The models were quantized using various schemes provided by the auto-round tool with MTP layers explicitly enabled. For multimodal use, projector files (mmproj) are provided in F16, BF16, and F32 formats.

Files and Sizes

File Name	Quant Type	Size	Description
`Qwen3.6-35B-A3B-Q2_K_S.gguf`	Q2_K_S	12.4 GB	Extremely high compression, significant quality loss.
`Qwen3.6-35B-A3B-Q2_K_MIXED.gguf`	Q2_K_MIXED	12.9 GB	Recommended high-compression option. Fast inference.
`Qwen3.6-35B-A3B-Q3_K_S.gguf`	Q3_K_S	15.5 GB	Very high compression, notable quality loss.
`Qwen3.6-35B-A3B-Q3_K_M.gguf`	Q3_K_M	16.4 GB	Balanced 3-bit quantization.
`Qwen3.6-35B-A3B-Q3_K_L.gguf`	Q3_K_L	18.6 GB	High quality 3-bit quantization.
`Qwen3.6-35B-A3B-Q4_0.gguf`	Q4_0	20.2 GB	Standard 4-bit quantization, good balance.
`Qwen3.6-35B-A3B-Q4_1.gguf`	Q4_1	22.4 GB	Higher quality 4-bit quantization than Q4_0.
`Qwen3.6-35B-A3B-Q4_K_S.gguf`	Q4_K_S	20.4 GB	Small 4-bit K-quant, good efficiency.
`Qwen3.6-35B-A3B-Q4_K_M.gguf`	Q4_K_M	21.7 GB	Recommended 4-bit K-quant, excellent balance.
`Qwen3.6-35B-A3B-Q5_0.gguf`	Q5_0	24.6 GB	Standard 5-bit quantization, very high quality.
`Qwen3.6-35B-A3B-Q5_1.gguf`	Q5_1	26.7 GB	Higher quality 5-bit quantization than Q5_0.
`Qwen3.6-35B-A3B-Q5_K_S.gguf`	Q5_K_S	24.6 GB	Small 5-bit K-quant, very high quality.
`Qwen3.6-35B-A3B-Q5_K_M.gguf`	Q5_K_M	25.3 GB	Recommended 5-bit K-quant, near-lossless.
`Qwen3.6-35B-A3B-Q6_K.gguf`	Q6_K	29.2 GB	6-bit K-quant, virtually indistinguishable from F16.
`Qwen3.6-35B-A3B-Q8_0.gguf`	Q8_0	37.8 GB	8-bit quantization, near-lossless.
`mmproj-model-f16.gguf`	F16	0.9 GB	Unified Projector in Float16 format.
`mmproj-model-bf16.gguf`	BF16	0.9 GB	Unified Projector in BFloat16 format.
`mmproj-model-f32.gguf`	F32	1.8 GB	Unified Projector in Float32 format.

Note: File sizes are slightly larger than non-MTP quants due to the additional MTP head weights.

Generate the Model

The models were generated using Intel's AutoRound with MTP layers explicitly enabled:

bash

auto-round \
    --model Qwen/Qwen3.6-35B-A3B \
    --output_dir ./quantized/ \
    --scheme <SCHEME> \
    --enable_alg_ext \
    --enable_torch_compile \
    --options '{"mtp_num_hidden_layers": 1, "num_nextn_predict_layers": 1}'

Usage with llama.cpp

These models can be used with a recent build of llama.cpp (must include Qwen3.5+ MTP support). For multimodal usage, specify the projector file:

./llama-cli -m Qwen3.6-35B-A3B-Q4_K_M.gguf --mmproj mmproj-model-f16.gguf --image your_image.jpg -p "Describe this image."

About AutoRound

AutoRound is an advanced quantization technique from Intel that aims to minimize accuracy loss through automated rounding optimization.

Support

These quantized models are made in my spare time using expensive hardware such as DGX Spark systems for quantization and validation. If you find these GGUFs useful for your projects, consider buying me a coffee to help cover hardware and compute costs. Every bit of support helps me keep producing high-quality quantized models for the community!

☕ Support me on Ko-fi

sphaela/Qwen3.6-35B-A3B-AutoRound-GGUF

Qwen3.6-35B-A3B GGUF (AutoRound Quantized, MTP Enabled)