OsaurusAI/Ornith-1.0-9B-MXFP4

Ornith-1.0-9B · MXFP4

Official OsaurusAI MXFP4 build of deepreinforce-ai/Ornith-1.0-9B (MIT) — a vision-language model on a Qwen3.5 hybrid backbone. Near-lossless 8-bit microscaled FP; runs on Apple Silicon via Osaurus / mlx.

• ~5.3 GB (from ~18.8 GB bf16) bundle.
• MXFP8: microscaled FP4 (group-size 32, 4-bit) on the language-model linear weights; the vision tower is preserved at fp16, short-conv kernels and norms kept fp16.
• Vision-language (image + text → text).

Architecture

Family	qwen3_5 (dense, hybrid)
Text layers	32 — 24 Gated-DeltaNet (linear-attention) + 8 full-attention
Hidden	4096 · untied lm_head
Vision	ViT tower (model.visual) preserved fp16
Cache	hybrid (GDN state + KV for attention layers)

Usage

# text
python -m mlx_lm generate --model OsaurusAI/Ornith-1.0-9B-MXFP4 --prompt "Explain a hash map in two sentences."

For image+text, load in Osaurus or an MLX-VLM runtime that supports qwen3_5 vision.

Provenance

• Base: deepreinforce-ai/Ornith-1.0-9B © DeepReinforce — MIT (Qwen3.5-based)
• Quantization: Osaurus · MXFP4 (microscaled FP4, group-size 32; vision tower fp16) · eric@osaurus.ai