Skip to content

gliclass:v4-edge

3 TagsUpdated 440M params1024 contextEnglishApache-2.0by Knowledgator

GLiClass v4 Edge (DeBERTa-v3-large backbone, a calibrator on every label's score), Apache-2.0: 0.506 on typed decisions.

440m
ollaya run gliclass:v4-edge --preset triage "I was charged twice for my subscription this month and want a refund."

Details

  • graph4c58ea35d7c3 · 5 MBonnx · deberta-v3-large · 440M · fp32
  • weights672cabebce49 · 882 MBhuggingface.co/knowledgator/gliclass-v4-edge/resolve/2486b93…/model.safetensors
  • tokenizerfac3bf8dbdea · 8 MBhuggingface.co/knowledgator/gliclass-v4-edge/resolve/2486b93…/tokenizer.json
  • decision605a7faa6991 · 1 KB{"engine": "onnx", "family": "gliclass", "encoder": "microsoft/deberta-v3-large", "layout": "gliclass-uni-v1", …}
  • calibration316915b3e458 · 84 B{"temperature": [1.0, 1.0, 1.0]}
  • license1b5e2bd856f6 · 10 KBGLiClass v4 Edge by Knowledgator (https://huggingface.co/knowledgator/gliclass-v4-edge)

Every layer is checked against its sha256 when it is pulled. Weights and tokenizers download from the model author's Hugging Face repository at a pinned commit; Ollaya never re-hosts them.

GLiClass readme and all models