# Can the Apple M4 (16 GB) run ZwZ2-4B?

> Yes. The Apple M4 (16 GB) runs ZwZ2-4B fully in unified memory at FP16 / BF16 (about 9.7 GB of 11 GB), at roughly 7.3 tokens/s.

Apple M4 (16 GB): 16 GB (10.7 GB usable), 120 GB/s. Estimates at an 8K context.

## Every quantisation

| Quant | Quality | ≈ GB | Fit | Est. tok/s |
|---|---|---|---|---|
| Q2_K | Heavy loss | 3.19 | fits | 44 |
| Q3_K_M | Noticeable loss | 3.53 | fits | 30 |
| Q4_K_M | Small loss | 4.15 | fits | 24 |
| Q5_K_M | Near-lossless | 4.72 | fits | 20 |
| Q6_K | Near-lossless | 5.11 | fits | 18 |
| Q8_0 | Lossless | 5.88 | fits | 14 |
| F16 (pick) | Lossless | 9.74 | fits | 7.3 |

## Run it

F16 straight from Hugging Face (ollama):

```bash
ollama run hf.co/inclusionAI/ZwZ2-4B-GGUF:F16
```

OpenAI-compatible server with a web UI (llama.cpp):

```bash
llama-server -hf inclusionAI/ZwZ2-4B-GGUF:F16
```

Download with the lms CLI (lmstudio):

```bash
lms get inclusionAI/ZwZ2-4B-GGUF@F16
```

Ask about an image (transformers):

```python
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="inclusionAI/ZwZ2-4B", dtype="auto", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])
```


## Vision models that fit comfortably

- [Qwen/Qwen3.5-4B](https://selfhost.directory/models/qwen/qwen3.5-4b/on/apple-m4-16gb): Q8_0 ≈5.88 GB, est. 13 tok/s
- [google/gemma-3-4b-it](https://selfhost.directory/models/google/gemma-3-4b-it/on/apple-m4-16gb): Q8_0 ≈5.2 GB, est. 14 tok/s
- [Qwen/Qwen3-VL-4B-Instruct](https://selfhost.directory/models/qwen/qwen3-vl-4b-instruct/on/apple-m4-16gb): Q8_0 ≈5.82 GB, est. 14 tok/s
- [google/medgemma-1.5-4b-it](https://selfhost.directory/models/google/medgemma-1.5-4b-it/on/apple-m4-16gb): Q8_0 ≈5.6 GB, est. 14 tok/s
- [Qwen/Qwen2.5-VL-3B-Instruct](https://selfhost.directory/models/qwen/qwen2.5-vl-3b-instruct/on/apple-m4-16gb): Q8_0 ≈3.99 GB, est. 17 tok/s
- [google/translategemma-4b-it](https://selfhost.directory/models/google/translategemma-4b-it/on/apple-m4-16gb): Q8_0 ≈5.6 GB, est. 12 tok/s

## Other GPUs that run it

- [GeForce RTX 4060](https://selfhost.directory/models/inclusionai/zwz2-4b/on/rtx-4060): Q8_0, est. 31 tok/s
- [Radeon RX 6700 XT](https://selfhost.directory/models/inclusionai/zwz2-4b/on/rx-6700-xt): F16, est. 23 tok/s
- [GeForce RTX 3060 12GB](https://selfhost.directory/models/inclusionai/zwz2-4b/on/rtx-3060): F16, est. 22 tok/s
- [Arc B580](https://selfhost.directory/models/inclusionai/zwz2-4b/on/arc-b580): F16, est. 28 tok/s
- [Tesla T4](https://selfhost.directory/models/inclusionai/zwz2-4b/on/tesla-t4): F16, est. 19 tok/s
- [Arc A770 16GB](https://selfhost.directory/models/inclusionai/zwz2-4b/on/arc-a770): F16, est. 34 tok/s
- [Radeon RX 7800 XT](https://selfhost.directory/models/inclusionai/zwz2-4b/on/rx-7800-xt): F16, est. 38 tok/s
- [Tesla P40](https://selfhost.directory/models/inclusionai/zwz2-4b/on/tesla-p40): F16, est. 21 tok/s

---
Model: https://selfhost.directory/models/inclusionai/zwz2-4b. GPU: https://selfhost.directory/models/gpus/apple-m4-16gb. Source: https://selfhost.directory/models/inclusionai/zwz2-4b/on/apple-m4-16gb