---
library_name: mlx
license: apache-2.0
pipeline_tag: text-generation
base_model: apodex/Apodex-1.1-mini
tags:
  - mlx
  - qwen3_5_moe
  - moe
  - 5-bit
---

# nicolasembleton/Apodex-1.1-mini-MLX-5bit

This model [nicolasembleton/Apodex-1.1-mini-MLX-5bit](https://huggingface.co/nicolasembleton/Apodex-1.1-mini-MLX-5bit) was converted to MLX format from [apodex/Apodex-1.1-mini](https://huggingface.co/apodex/Apodex-1.1-mini) using mlx-lm version **0.31.3**.

Qwen3.5-35B-A3B MoE, 36B params / 3B active.

| Property | Value |
|----------|-------|
| Quantization | 5-bit affine |
| Bits per weight | 5.502 |
| Group size | 64 |
| Size | 23.9 GB |

## Use with mlx

```bash
pip install mlx-lm
mlx_lm.generate --model nicolasembleton/Apodex-1.1-mini-MLX-5bit --prompt "Hello"
```

```python
from mlx_lm import load, generate

model, tokenizer = load("nicolasembleton/Apodex-1.1-mini-MLX-5bit")

prompt = "Hello"

if tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=True)
```

## Architecture

- **Base:** Qwen3.5-35B-A3B (MoE)
- **Experts:** 256 total, 8 active per token
- **Layers:** 40 (hybrid linear/full attention)
- **Context:** 262144 max position embeddings
- **Vision tower:** Stripped during conversion (text-only)

## Known Issues

Metal OOM during long-context prefill on less than 48 GB unified memory. Keep context under 32k tokens.