Qwen3-VL-2B-Instruct-MP-GGUF

Mixed Precision GGUF layer quantization of Qwen3-VL-2B-Instruct by Qwen

Original model: https://huggingface.co/Qwen/Qwen3-VL-2B-Instruct

The hybrid quant employs different quantization levels on a per layer basis to enable both high performance and small file size at the same time. This quant was optimized for high performance across a set of test prompts with ~Q8_0 size. The model predominantly exhibits rep fails across a set of curated test prompts, where it falls into infinite repeat loops on most prompts when using greedy sampling. Extensive testing showed there is no way to correct this problem by adjusting the layer quants, the problem is baked into the model by the training process. The model is still usable with simple vision prompts but will often rep fail if asked to solve a prompt with step by step reasoning with greedy sampling. The VL 32B Instruct model does not exhibit this failure mode.

The quants employed are all K to avoid slow CPU or older GPU processing of IQ quants. For this file the layer quants are as follows:

Q6_K_S : Q6_K
Q6_K_M : attn_v = q8_0 ffn_d = q8_0
Q6_K_L : attn_v = q8_0 attn_o = q8_0 ffn_d = q8_0

LAYER_TYPES='[
   [0 ,"Q8_0"  ],[1 ,"Q8_0"  ],[2 ,"Q8_0"  ],[3 ,"Q8_0"  ],[4 ,"Q6_K_L"],[5 ,"Q6_K_L"],
   [6 ,"Q6_K_L"],[7 ,"Q6_K_L"],[8, "Q6_K_L"],[9, "Q6_K_L"],[10,"Q6_K_M"],[11,"Q6_K_M"],
   [12,"Q6_K_S"],[13,"Q5_K_M"],[14,"Q5_K_M"],[15,"Q6_K_S"],[16,"Q6_K_M"],[17,"Q6_K_M"],
   [18,"Q6_K_L"],[19,"Q6_K_L"],[20,"Q6_K_L"],[21,"Q6_K_L"],[22,"Q6_K_L"],[23,"Q6_K_L"],
   [24,"Q8_0"  ],[25,"Q8_0"  ],[26,"Q8_0"  ],[27,"Q8_0"  ]
   ]'
   FLAGS="--token-embedding-type Q8_0 --output-tensor-type Q8_0 --layer-types-high"

Comparison:

Usage:

Qwen3-VL-2B-Instruct is a vision capable model. It can be used together with its multimedia projector layers to process images and text inputs and generate text outputs. The mmproj file is made available in this repository. To test vision mode follow the docs in the mtmd readme in the tools directory of the source tree https://github.com/ggml-org/llama.cpp/blob/master/tools/mtmd/README.md .

On a 4070 non-code gen rate is about 185tps.

Llama.cpp minimum version to run Qwen3-VL series should be 6915 with recommended 6936 and above.

Benchmarks:

A full set of vision benchmarks for the model is given here: https://huggingface.co/spaces/steampunque/benchlm

Download the file from below:

A discussion thread about the hybrid layer quant approach can be found here on the llama.cpp git repository:

https://github.com/ggml-org/llama.cpp/discussions/13040