Skip to content

kev:4b-fp16

6 TagsUpdated 4.2B params8192 contextEnglishApache-2.0by Jared Palmer

Kev on Qwen3.5-4B base (LoRA and pointer head): more accurate than 0.8b out of domain. Needs about 9 GB of memory. This tag computes in fp16 on NVIDIA GPUs: faster than kev:4b, with at least 99.5 % of its decisions the same. On a CPU it runs, slowly; use kev:4b there.

4.2b
ollaya run kev:4b-fp16 --preset triage "I was charged twice for my subscription this month and want a refund."

Details

  • graph504883970526 · 15 MBonnx · kev · 4.2B · fp16
  • weightsdf547074dce7 · 5.3 GBhuggingface.co/Qwen/Qwen3.5-4B-Base/resolve/1001bb4…/model.safetensors-00001-of-00002.safetensors
  • weights590fbaac095d · 4.0 GBhuggingface.co/Qwen/Qwen3.5-4B-Base/resolve/1001bb4…/model.safetensors-00002-of-00002.safetensors
  • weights90e817356246 · 130 MBhuggingface.co/jaredpalmer/kev-4b/resolve/139fdd9…/adapter_model.safetensors
  • weightsdd633435998e · 5 MBhuggingface.co/jaredpalmer/kev-4b/resolve/139fdd9…/head.pt
  • tokenizer06b9509352d2 · 20 MBhuggingface.co/jaredpalmer/kev-4b/resolve/139fdd9…/tokenizer.json
  • decision27e748d8c47f · 3 KB{"engine": "onnx", "family": "kev", "encoder": "", "layout": "kev-pointer-v1", …}
  • calibrationa51c52e256db · 252 B{"temperature": [2.406, 2.406, 2.406]}
  • license5ddbf0af0c82 · 10 KBKev by Jared Palmer (https://huggingface.co/jaredpalmer/kev-4b)

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.

Kev readme and all models