Skip to content

coba:68m-ternary-fp16

16 TagsUpdated 75M params512 contextEnglishApache-2.0by Ollaya

The smallest: ternary Ettin 68M, in-domain.

fine-tuned75m
ollaya run coba:68m-ternary-fp16 --preset triage "I was charged twice for my subscription this month and want a refund."

Details

  • graphfb4e9fcb360e · 2 MBonnx · encoder · 75M · fp32
  • graph59bbf665d7e0 · 3 MBonnx · encoder · 75M · fp16
  • weights65dbde4c16fb · 299 MBhuggingface.co/ollaya-dev/coba-68m-ternary/resolve/e0c9c8f…/model.safetensors
  • tokenizer6c8aaa9a5420 · 4 MBhuggingface.co/ollaya-dev/coba-68m-ternary/resolve/e0c9c8f…/tokenizer/tokenizer.json
  • decisionb01dafbc1f88 · 457 B{"engine": "onnx", "family": "laya", "encoder": "encoder", "layout": "laya-markers-v1", …}
  • calibrationf4a8b5dd4e1c · 87 B{"temperature": [0.96, 0.83, 0.79]}
  • licenseef14791048a7 · 10 KBCoba by Ollaya (https://huggingface.co/ollaya-dev)
  • archa6f8e9f4c2c5 · 1 KBapplication/vnd.ollaya.arch
  • paramsa1d447ba672f · 21 Bprecision fp16

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.

Variant of

coba readme and all models