No — not on this device
Ornith 1.5 397B-A17B at Q4_K_M needs 224.1 GB against 24.0 GB usable, and the shortfall of 200.1 GB is more than 256 GB of system RAM can cover at a tolerable speed. A smaller sibling or a lower quantisation is the honest answer here.
The flagship Ornith on the Qwen3.5 397B-A17B architecture, MIT-licensed. A 256 GB Mac Studio at Q4, and it is in the Ollama library.
What hardware do I need for Ornith 1.5 397B-A17B? →
Fits instead: Ornith 1.5 35B-A3B (20.9 GB) · Ornith 1.5 9B (6.1 GB)
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 392.8 GB | 393.7 GB | — | ~3.1 | −0.1% ppl | 369.7 GB over |
| Q6_K | 303.2 GB | 304.0 GB | — | ~4.0 | −0.4% ppl | 280.0 GB over |
| Q5_K_M | 262.0 GB | 262.9 GB | — | ~4.6 | −0.8% ppl | 238.9 GB over |
| Q4_K_M | 223.2 GB | 224.1 GB | — | ~5.5 | −1.9% ppl | 200.1 GB over |
| Q3_K_M | 180.7 GB | 181.5 GB | — | ~6.7 | −5.4% ppl | 157.5 GB over |
| Q2_K | 154.8 GB | 155.7 GB | — | ~7.9 | −15% ppl | 131.7 GB over |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 15 of its 60 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/Ornith-1.5-397B-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.