README.md

---
license: apache-2.0
base_model: mistralai/Devstral-Small-2-24B-Instruct-2512
tags:
  - lora
  - peft
  - mlx
  - eu-kiki
  - eu-ai-act
language:
  - fr
  - en
library_name: peft
---

# eu-kiki-devstral-cpp-lora

LoRA adapter for **mistralai/Devstral-Small-2-24B-Instruct-2512**, part of the [eu-kiki](https://github.com/L-electron-Rare/eu-kiki) project — a 100 % EU-sovereign multi-model LLM serving pipeline. EU AI Act Article 52/53 compliant.

## Performance

**HumanEval (custom Studio scorer, EvalPlus extra-tests not run):** base 87.20 → +cpp 85.98 = −1.22 pts.

## Usage

```python
from mlx_lm import load
from mlx_lm.tuner.utils import linear_to_lora_layers
from huggingface_hub import snapshot_download

base_path = snapshot_download("mistralai/Devstral-Small-2-24B-Instruct-2512")
adapter_path = snapshot_download("clemsail/eu-kiki-devstral-cpp-lora")

model, tokenizer = load(base_path)
linear_to_lora_layers(model, num_layers=32, config={"rank": 16, "alpha": 32})
model.load_weights(f"{adapter_path}/adapters.safetensors", strict=False)
```

Or, simpler, fuse and serve via `mlx_lm fuse`:

```bash
python -m mlx_lm fuse \
    --model mistralai/Devstral-Small-2-24B-Instruct-2512 \
    --adapter-path <adapter_path> \
    --save-path /tmp/eu-kiki-devstral-cpp-lora-fused \
    --dequantize
```

## Training configuration

| Parameter | Value |
|---|---|
| Method | LoRA |
| Rank | 16 |
| Alpha | 32 |
| Dropout | 0.05 |
| Target modules | q_proj, k_proj, v_proj, o_proj |
| Precision | BF16 |
| Optimiser | AdamW |
| Learning rate | 1e-5 |
| Framework | MLX (`mlx_lm` fork on Apple Silicon) |
| Hardware | Mac Studio M3 Ultra 512 GB unified memory |

## Provenance & EU AI Act compliance

Datasets used to train this adapter are HF-traceable. Per-source SPDX licenses, download dates, source row counts, and used row counts are documented in:

- [`docs/eu-ai-act-transparency.md`](https://github.com/L-electron-Rare/eu-kiki/blob/main/docs/eu-ai-act-transparency.md) — system-level transparency record (Art. 52/53)
- [`MODEL_CARD.md`](https://github.com/L-electron-Rare/eu-kiki/blob/main/MODEL_CARD.md) — full evaluation summary across HumanEval+, MT-Bench, GSM8K, KIKI-DSL v3
- [`eval/results/SUMMARY.md`](https://github.com/L-electron-Rare/eu-kiki/blob/main/eval/results/SUMMARY.md) — per-bench reproducible results

## Risk classification

**Limited risk** (EU AI Act Art. 52). General-purpose AI; not deployed in safety-critical contexts.

## License

Apache 2.0, matching the base model.

## Citation

```bibtex
@misc{eu-kiki-2026,
  title  = {eu-kiki: EU-sovereign multi-model LLM serving with HF-traceable LoRA adapters},
  author = {Saillant, Clément},
  year   = {2026},
  url    = {https://github.com/L-electron-Rare/eu-kiki},
  note   = {Live demo: https://ml.saillant.cc}
}
```