Ailiance-fr/devstral-cpp-lora

Ailiance — Devstral-Small-2-24B-Instruct cpp LoRA

LoRA adapter fine-tuned on mistralai/Devstral-Small-2-24B-Instruct-2512 for cpp tasks.

Maintained by Ailiance — French AI org publishing EU AI Act aligned LoRA adapters and datasets.

Quick start (MLX)

from mlx_lm import load, generate

model, tokenizer = load(
    "mistralai/Devstral-Small-2-24B-Instruct-2512",
    adapter_path="Ailiance-fr/devstral-cpp-lora",
)

print(generate(model, tokenizer, prompt="..."))

Training

Hyperparameter	Value
Base model	mistralai/Devstral-Small-2-24B-Instruct-2512
Method	LoRA via mlx-lm
Rank	16
Scale	2.0
Alpha	32
Max seq length	2048
Iterations	500
Optimizer	Adam, LR 1e-5
Hardware	Apple M3 Ultra 512 GB

Training data lineage

Derived from the internal eu-kiki / mascarade curation. All upstream samples are synthetic, permissively-licensed, or generated from Apache-2.0 base resources. See the Ailiance-fr catalog for related cards.

Training metrics

Extracted from training log (batch_eu_kiki_v2.log):

Metric	Value
Final train loss	0.603
Final validation loss	0.401
Val loss reduction	+1.779 (from 2.180)
Iterations completed	500
Trainable parameters	0.224% (279.708M / 125025.989M)

Validation loss is measured every 200 iterations on a held-out split of the training corpus (val_batches=5, mlx-lm LoRA trainer).

Benchmark on production tasks

This LoRA has not yet been evaluated through the electron-bench functional benchmark pipeline. The current pipeline targets the gemma-4-E4B base only; support for the devstral base is on the roadmap (open issues).

For a comparable reference matrix on a related domain (electronics, embedded, KiCad), see the Gemma champions:

Adapter	Highlights
Ailiance-fr/gemma-4-E4B-eukiki-lora	+55 P1-DSL, +42 P1-PCB, +25 SPICE, +38 P3
Ailiance-fr/gemma-4-E4B-mascarade-lora	+48 P3 extraction

Full base-vs-LoRA matrix: compare_base_vs_lora.md.

License chain

Component	License
Base model (mistralai/Devstral-Small-2-24B-Instruct-2512)	apache-2.0
Training data (internal Ailiance curation (synthetic + permissive sources))	apache-2.0
LoRA adapter (this repo)	apache-2.0

All upstream components are Apache 2.0 / MIT — LoRA inherits permissive terms.

EU AI Act compliance

• Article 53(1)(c): training data licenses preserved (per-dataset cards declare upstream licenses).
• Article 53(1)(d): training data summary — see upstream dataset cards on Ailiance-fr.
• GPAI Code of Practice (July 2025): base mistralai/Devstral-Small-2-24B-Instruct-2512 released under apache-2.0.
• No web scraping by Ailiance, no licensed data, no PII.
• Upstream Stack Exchange content (where applicable) is CC-BY-SA-4.0 and propagates to this adapter.

License

LoRA weights: apache-2.0 — see License chain table above for derivation rationale.

Citation

@misc{ailiance_devstral_cpp_2026,
  author    = {Ailiance},
  title     = {Ailiance — Devstral-Small-2-24B-Instruct cpp LoRA},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/Ailiance-fr/devstral-cpp-lora}
}

Related

See the full Ailiance-fr LoRA collection.

Bench comparison (2026-05-11)

Base model (Devstral-Small-2-24B-MLX-4bit) capability

Task	Score	Notes
GSM8K-CoT flex EM	0.96	W3 lm-eval-harness (--limit 100)
ARC-Easy acc / acc_norm	0.80 / 0.75
MMLU-Pro Computer Science	0.64

Source: https://github.com/ailiance/ailiance/tree/main/output/lm-eval-base-2026-05-11

This LoRA (tuned) — bench PENDING

Will include kicad-sch / iact-bench validators + W3 lm-eval delta. See spec for methodology: https://github.com/ailiance/ailiance-bench/blob/main/docs/superpowers/specs/2026-05-11-kicad-sch-gap-design.md

Upstream base model — official evaluations

This LoRA fine-tunes mistralai/Devstral-Small-2-24B-Instruct-2512, Mistral's coding-specialist LLM. Headline software-engineering benchmarks from the upstream model card:

Benchmark	Devstral Small 2 (24B)	Devstral 2 (123B)	DeepSeek v3.2 (671B)	Claude Sonnet 4.5
SWE Bench Verified	68.0 %	72.2 %	73.1 %	77.2 %
SWE Bench Multilingual	55.7 %	61.3 %	70.2 %	68.0 %
Terminal Bench 2	22.5 %	32.6 %	46.4 %	42.8 %

(For reference, GPT-5.1 Codex High: 73.7 % SWE Verified · 52.8 % Terminal Bench 2.)

Devstral Small 2 (24B) is competitive with much larger open models on SWE Bench Verified (e.g. matches GLM-4.6 at 355B). Architecture uses rope-scaling per Llama 4 + Scalable-Softmax (arXiv:2501.19399).

Source: official Devstral-Small-2-24B-Instruct-2512 model card.

Reading these alongside this LoRA: Devstral Small 2 is a strong coding base. This LoRA inherits its SWE-Bench performance and adds language- or domain-specific specialization.