qwen3-0.6b-queues-university-classifier-de

qwen3-0.6b-queues-university-classifier-de

A 20-way text classifier for routing German university support tickets to the correct queue. Fine-tuned from Qwen/Qwen3-0.6B-Base as Qwen3ForSequenceClassification on the queues-university synthetic dataset (~11.4k examples).

Intended use

Classify a ticket (subject + body, concatenated as the input field) into one of 20 canonical university queues:

BEWERBUNG_ZULASSUNG, IMMATRIKULATION_STATUS, PRUEFUNGEN_NOTEN, STUDIENBERATUNG_ORIENTIERUNG, LEHRE_MODULE, IT_CAMPUS_SYSTEME, BIBLIOTHEK_MEDIEN, INTERNATIONAL_MOBILITY, FORSCHUNG_DRITTMITTEL, TRANSFER_PATENTE, LABORE_SICHERHEIT, GEBAEUDE_RAEUME, FINANZEN_GEBUEHREN, STIPENDIEN_SOZIALES, WOHNEN_CAMPUSLEBEN, PERSONAL_HR, GLEICHSTELLUNG_BARRIEREFREIHEIT, CAREER_PRAKTIKA, SPORT_GESUNDHEIT, RECHT_DATENSCHUTZ.

Quickstart

from transformers import pipeline

clf = pipeline(
    "text-classification",
    model="open-ticket-ai/qwen3-0.6b-queues-university-classifier-de",
)
clf("Mein Login zum Campus-Portal funktioniert nicht, ich habe mein Passwort vergessen.")
# -> [{'label': 'IT_CAMPUS_SYSTEME', 'score': 0.98}]

Training

• Base: Qwen/Qwen3-0.6B-Base
• Task head: AutoModelForSequenceClassification (20 labels)
• Epochs: 1, max_seq_length 256, batch 8, grad_accum 2, lr 2e-5, bf16
• Hardware: 1x A100 (Vertex AI europe-west4)
• Training script: otai-studio/apps/cloud_backend/src/ai/training/scripts/train_seq_cls.py

Limitations

• Dataset is synthetic German university support tickets — distribution may differ from your tenant.
• Not safety-aligned; do not use for content moderation.