Ailiance-fr/apertus-emc-dsp-power-curriculum-lora

Ailiance — Apertus-70B-Instruct emc-dsp-power (curriculum) LoRA

LoRA adapter fine-tuned on swiss-ai/Apertus-70B-Instruct-2509 for emc-dsp-power tasks.

Variant: trained with multi-phase length curriculum.

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

## ✅ Training data attribution audited (2026-05-11) Trained on `Ailiance-fr/mascarade-emc-dataset` (4.05% SE), `Ailiance-fr/mascarade-dsp-dataset` (5.35% SE), `Ailiance-fr/mascarade-power-dataset` (4.87% SE). Per-sample Stack Exchange Electronics attribution recovered via SE API search: only ~4.05–5.35 % of samples per dataset are from Stack Exchange Electronics (now fully attributed in metadata.stack_exchange_attribution — URL + author + postid + creationdate). Original heuristic estimate of "~30 % SE" was over-counted by ~6–7× (style ≠ source). See audit trail for the methodology (/search/advanced + body match ≥ 0.60). Remaining content is either (a) style-resembling SE but not findable on the API (marked attribution_recovery=not_found_on_se, probable synthetic) or (b) synthetic LLM-generated.

Quick start (MLX)

from mlx_lm import load, generate

model, tokenizer = load(
    "swiss-ai/Apertus-70B-Instruct-2509",
    adapter_path="Ailiance-fr/apertus-emc-dsp-power-curriculum-lora",
)

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

Training

Training data lineage

For per-sample provenance and attribution status, consult the dataset card.

Benchmark roadmap

This LoRA has not yet been evaluated through electron-bench (the current pipeline supports gemma-4-E4B base only). Training was completed with the standard mlx-lm LoRA trainer (rank 16, alpha 32, scale 2.0, AdamW LR 1e-5, 500 iters) — full hyperparameters are in the Training table above.

Planned evaluations:

• —Perplexity on the validation split of the training data
• —Functional benchmark on apertus-specific tasks
• —Comparison vs base swiss-ai/Apertus-70B-Instruct-2509

Track progress: ailiance-bench issues.

For reference benchmarks on the gemma-4-E4B base, see the base-vs-LoRA matrix.

License chain

Most restrictive license in the chain (CC-BY-SA-4.0 share-alike) propagates to derivatives.

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 swiss-ai/Apertus-70B-Instruct-2509 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: cc-by-sa-4.0 — see License chain table above for derivation rationale.

Citation

@misc{ailiance_apertus_emc_dsp_power_curriculum_2026,
  author    = {Ailiance},
  title     = {Ailiance — Apertus-70B-Instruct emc-dsp-power (curriculum) LoRA},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/Ailiance-fr/apertus-emc-dsp-power-curriculum-lora}
}

Related

See the full Ailiance-fr LoRA collection.

Bench comparison (2026-05-11)

Base model (Apertus-70B-Instruct-2509) capability

This LoRA (tuned) — bench PENDING

Production usage: served via gateway alias ailiance-apertus-<domain> on <https://www.ailiance.fr> through the Apertus multi-LoRA hot-swap server (Studio :9322, 1 base + 10 LoRA dynamic swap, ~40GB VRAM).