Qwen3.5-2B-Agent GGUF

GGUF conversion of Ma7ee7/Qwen3.5-2B-Agent for

llama.cpp-compatible applications.

The multimodal projector is included as mmproj-Qwen3.5-2B-Agent-Q8_0.gguf.

Converted with llama.cpp commit 720d7fa4097f76e5d0eade5a92c1df87c1faf9d9.

Files

| File | Size | Description |

|---|---:|---|

| Qwen3.5-2B-Agent-BF16.gguf | 3.63 GiB | High-fidelity source GGUF for archival use or requantization. |

| Qwen3.5-2B-Agent-Q8_0.gguf | 1.93 GiB | Near-original quality with moderate compression. |

| Qwen3.5-2B-Agent-Q6_K.gguf | 1.50 GiB | Very high quality with lower memory use than Q8_0. |

| Qwen3.5-2B-Agent-Q5_K_M.gguf | 1.35 GiB | Strong quality-to-size balance. |

| Qwen3.5-2B-Agent-Q4_K_M.gguf | 1.22 GiB | Recommended default for most local inference. |

| mmproj-Qwen3.5-2B-Agent-Q8_0.gguf | 347.77 MiB | Multimodal projector required for image input. |

Recommendation

Start with Q4_K_M. Use Q5_K_M or Q6_K when you have more memory and want

higher fidelity. Q8_0 is the largest quantized option in this repository.

llama.cpp

llama-cli -hf Ma7ee7/Qwen3.5-2B-Agent-GGUF:Q4_K_M -cnv --jinja

For image input, download the matching mmproj-*.gguf file and pass it with

--mmproj.