Ailiance-fr/router-v5-minilm

ailiance-router-v5 (MiniLM L6 v2 + 2-layer MLP)

Domain classifier head used by the ailiance gateway to route incoming chat requests to the most adapted backend worker.

Performance

• Top-1 accuracy: 87.6 %
• Top-3 accuracy: 98.7 %
• Validation set: 1 978 prompts across 32 domains.

Architecture

• Encoder: sentence-transformers/all-MiniLM-L6-v2 (frozen, 384d)
• Head: 2-layer MLP Linear(384→256) -> GELU -> Dropout(0.1) -> Linear(256→32)
• Loss: BCE-with-logits (multi-label sigmoid)
• Optimizer: AdamW lr=1e-3, batch=128, 30 epochs, seed=42

🇪🇺 EU AI Act transparency

This artefact is published under the AI Act framework (Regulation EU 2024/1689).

Field	Value
Provider	Ailiance (clemsail)
Role under AI Act	GPAI provider (this head). Encoder is upstream Microsoft / sentence-transformers under Apache-2.0.
Adapter type	Custom 2-layer MLP head (~85 k parameters)
Base model	sentence-transformers/all-MiniLM-L6-v2
License	Apache-2.0
Intended use	Domain routing only — classify a user prompt over 32 technical domains, decide which downstream worker handles the request. Not for standalone classification of safety-critical text.
Out of scope	Healthcare triage, legal classification, autonomous safety decisions, anything requiring calibrated probabilities.
Risk classification	Limited risk — Article 50 obligations apply.
Training data origin	All open permissive-licensed datasets + Ailiance in-house curation. Full per-source provenance: https://ailiance.fr/transparency/router-provenance
Reproducibility	https://ailiance.fr/transparency/router-rebuild-recipe
Contact	postmaster@saillant.cc

Training data sources

Domain group	Source dataset	License	Rows
python, rust, ts, cpp, shell, html-css, security, devops	sahil2801/CodeAlpaca-20k	CC-BY-4.0	~3 700
sql	b-mc2/sql-create-context	CC-BY-4.0	805
chat-fr	OpenAssistant/oasst1 (FR-filtered)	Apache-2.0	801
math	openai/gsm8k	MIT	803
reasoning	AI-MO/NuminaMath-CoT	Apache-2.0	803
spice, power, electronics	theprint/Electronics-QA	Apache-2.0	~1 500
freecad	Yas1n/FreeCAD_Sketches (filename-derived)	CC-BY-4.0	602
lua-upy	Roblox/luau_corpus	MIT	603
calcul-normatif, docker, FR/EN augment, greetings	Ailiance in-house	Apache-2.0	528 + 200

Total clean corpus: 9 817 prompts across 32 domains.

Files

• router.safetensors — MLP head weights
• meta.json — encoder name, dim, head config, domain list
• label_map.json — domain ↔ index

Validated in ailiance/ailiance-bench v0.2

This model is referenced in the Ailiance benchmark suite (Phase 6 scoreboard, 7-task hardware-design evaluation).

See the full scoreboard: ailiance-bench README#scoreboard-lora-phase-6.