Skip to content

coba:400m-ternary-fp32

16 TagsUpdated 421M params512 contextEnglishApache-2.0by Ollaya

The 400m with ternary weights, in-domain: trained on the train splits of ten public benchmark tasks.

fine-tuned421m
ollaya run coba:400m-ternary-fp32 --preset triage "I was charged twice for my subscription this month and want a refund."

Details

  • graph5127480db99f · 3 MBonnx · encoder · 421M · fp32
  • graph40b093e6ac84 · 4 MBonnx · encoder · 421M · fp16
  • weights691ebad83f9e · 1.7 GBhuggingface.co/ollaya-dev/coba-400m-ternary/resolve/fd90709…/model.safetensors
  • tokenizer6c8aaa9a5420 · 4 MBhuggingface.co/ollaya-dev/coba-400m-ternary/resolve/fd90709…/tokenizer/tokenizer.json
  • decisionb01dafbc1f88 · 457 B{"engine": "onnx", "family": "laya", "encoder": "encoder", "layout": "laya-markers-v1", …}
  • calibration2da19eca5f3f · 87 B{"temperature": [1.03, 0.81, 0.76]}
  • licenseef14791048a7 · 10 KBCoba by Ollaya (https://huggingface.co/ollaya-dev)
  • arch8be02e5f6a3b · 1 KBapplication/vnd.ollaya.arch
  • paramsdd15c90375b8 · 21 Bprecision fp32

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