---

license: apache-2.0

base_model: Kolyadual/Newton-bot-3-VLM-mini-4B

tags:

• gguf
• qwen3.5
• vlm
• mamba
• llama.cpp

---

Newton-bot-3-VLM-mini-4B-GGUF

Full set of GGUF quantizations for Kolyadual/Newton-bot-3-VLM-mini-4B, based on the hybrid Qwen 3.5 architecture (Mamba + Attention).

⚠️ Vision / VLM Usage

This is a multimodal model. To process images, you must use the unquantized visual projector (mmproj-f16.gguf) alongside the quantized text models.

When using llama.cpp (via llama-server, llama-cli, or LM Studio), pass the F16 projector using the --mmproj flag.

Files

| File | Quantization | Type | Approx. Size |

|---|---|---|---|

| mmproj-f16.gguf | F16 | Visual Projector | 641 MB |

| newton-q2_k.gguf | Q2_K | Text Model | 1.8 GB |

| newton-q3_k_s.gguf | Q3_K_S | Text Model | 2.0 GB |

| newton-q3_k_m.gguf | Q3_K_M | Text Model | 2.2 GB |

| newton-q3_k_l.gguf | Q3_K_L | Text Model | 2.3 GB |

| newton-q4_0.gguf | Q4_0 | Text Model | 2.4 GB |

| newton-q4_1.gguf | Q4_1 | Text Model | 2.6 GB |

| newton-q4_k_s.gguf | Q4_K_S | Text Model | 2.5 GB |

| newton-q4_k_m.gguf | Q4_K_M | Text Model | 2.6 GB |

| newton-q5_0.gguf | Q5_0 | Text Model | 2.9 GB |

| newton-q5_1.gguf | Q5_1 | Text Model | 3.1 GB |

| newton-q5_k_s.gguf | Q5_K_S | Text Model | 2.9 GB |

| newton-q5_k_m.gguf | Q5_K_M | Text Model | 2.9 GB |

| newton-q6_k.gguf | Q6_K | Text Model | 3.3 GB |

| newton-q8_0.gguf | Q8_0 | Text Model | 4.3 GB |

| newton-f16.gguf | F16 | Text Model | 8.1 GB |

How to run locally

Example using llama-server:

llama-server \
  -m newton-q4_k_m.gguf \
  --mmproj mmproj-f16.gguf \
  -ngl 999 \
  --host 0.0.0.0 --port 8080