---

language:

• en

tags:

• qwen
• qwen3.5
• moe
• mxfp4
• gguf
• vision
• multimodal
• 35b

license: mit

base_model: deepreinforce-ai/Ornith-1.0-35B

---

Ornith 1.0 35B — MXFP4 GGUF

MXFP4 quantization of deepreinforce-ai/Ornith-1.0-35B, a 35B parameter Qwen3.5 MoE model with 256 experts (8 active per token) and vision support.

About the Model

Ornith 1.0 is a Qwen3.5 MoE architecture with:

• 35B total parameters with 8B active per token (256 experts, 8 active)
• 40-layer MoE decoder with sliding + full attention hybrid
• 27-layer vision encoder for multimodal image understanding
• 262K context window
• MIT License

Architecture

• Text model: Qwen3.5 MoE — 40 layers, 2048 hidden, 256 experts (8 active/token)
• Vision encoder: 27-layer SigLIP-style, 1152 hidden, patch_size 16
• Vocabulary: 248,320 tokens

Quantization

Quantized from the BF16 safetensors using llama.cpp (build 537).

MXFP4 (Microscaling FP4) uses block-wise quantization with shared exponents.

Files

| File | Size | Description |

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

| ornith-1.0-35b-mxfp4.gguf | ~17.4 GB | MXFP4 quantized model weights |

| mmproj-ornith-1.0-35b-f16.gguf | ~0.88 GB | Vision projector (BF16) |

Usage

llama-server \
  -m ornith-1.0-35b-mxfp4.gguf \
  --mmproj mmproj-ornith-1.0-35b-f16.gguf \
  -ngl 99 \
  --host 0.0.0.0 \
  --port 8080

Hardware Requirements

• Minimum: 20 GB VRAM
• Recommended: 24+ GB VRAM for full GPU offload

License

MIT