---
library_name: mlx
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/main/LICENSE
language:
- en
pipeline_tag: text-generation
base_model: Qwen/Qwen2.5-7B-Instruct
tags:
- chat
- mlx
- quantization
- bias-evaluation
- q3
---

# qwen2.5-7b-instruct-q3 (MLX, CBA artifact)

MLX-format 3-bit (Q3) variant of [`Qwen/Qwen2.5-7B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct).

This is one of the **15 model artifacts** from the paper:

> **Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels**
> Plawan Kumar Rath, Rahul Maliakkal. *IEEE Cloud Summit 2026*.
> Code: <https://github.com/plawanrath/compression-bias-amplification>
> arXiv: <https://arxiv.org/abs/2605.15208>

## Quantization

Weight-only post-training quantization via `mlx_lm.convert`:

- **bits:** 3
- **group_size:** 64
- **mode:** affine


## How this artifact was produced

```bash
python -m mlx_lm.convert \
    --hf-path Qwen/Qwen2.5-7B-Instruct \
    --mlx-path ./qwen2.5-7b-instruct-q3 \
    --quantize \
    --q-bits 3 \
    --q-group-size 64
```

This is the **exact** artifact used to produce the inference results in §4.3 of the paper (911,100 records over BBQ ambiguous, 5 seeds × 12,148 items × 15 configs).

## Usage (MLX)

```bash
pip install mlx-lm
```

```python
from mlx_lm import load, generate

model, tokenizer = load("plawanrath/qwen2.5-7b-instruct-q3-mlx-cba")
prompt = tokenizer.apply_chat_template(
    [{"role": "user", "content": "Hello!"}],
    add_generation_prompt=True,
    tokenize=False,
)
print(generate(model, tokenizer, prompt=prompt, max_tokens=128))
```

Or via CLI:

```bash
mlx_lm.generate --model plawanrath/qwen2.5-7b-instruct-q3-mlx-cba --prompt "Hello!"
```

## Paper findings relevant to this variant

The paper documents a **dose-response** relationship between quantization aggressiveness and emergent stereotypical behavior on BBQ ambiguous questions:

| Variant | % of BF16-unbiased items that became biased |
|---|---|
| Q8 | 0.1–0.9% |
| Q6 | 0.3–1.3% |
| Q4 | 2.2–5.6% |
| Q3 | 6.0–21.1% |

These changes are largely **invisible to perplexity** (<0.5% shift at Q8, <3% at Q4 across all three families). Treat any deployment of compressed instruction-tuned models on fairness-sensitive tasks accordingly.

## Model details

- **Base model:** [`Qwen/Qwen2.5-7B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct)
- **Family:** Qwen2
- **Parameters:** 7.6B
- **Precision:** 3-bit (Q3)
- **Format:** MLX (Apple Silicon)
- **Conversion framework:** [`mlx-lm`](https://github.com/ml-explore/mlx-lm)

## License

Inherited from the base model (`apache-2.0`). See the upstream model page for the full license text.

## Citation

```bibtex
@inproceedings{rath2026quantization,
  title     = { Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels },
  author    = {Rath, Plawan Kumar and Maliakkal, Rahul},
  booktitle = { IEEE Cloud Summit 2026 },
  year      = {2026},
  eprint    = {2605.15208},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2605.15208}
}
```