Skip to content

jevk5:4b

2 TagsUpdated 4B params16384 contextEnglishApache-2.0by alibiserikbay

JevK5 v0.3 (4B), Q8_0 GGUF, with the author's temperature 1.22: 0.625 on typed decisions; up to 16 options per question.

4b
ollaya run jevk5:4b --preset triage "I was charged twice for my subscription this month and want a refund."

Details

  • weightsaea433883bc7 · 4.5 GBgguf · Q8_0 · huggingface.co/alibiserikbay/JevK5-GGUF/resolve/ec67b0b…/jevk5-4b-v0.3-Q8_0.gguf
  • decision84abdb80a5d8 · 1 KB{"engine": "llama", "family": "jevk5", "layout": "jevk5-v1", "gguf": {…}, …}
  • calibration1940514cbbfc · 300 B{"temperature": [1.22, 1.22, 1.22]}
  • license8e946530c46d · 10 KBJevK5 by the JevK5 authors (https://github.com/allebee, https://huggingface.co/alibiserikbay), a fine-tune of Qwen3.5-4B by the Qwen team (Apache-2.0). Its decision prompt and one-pass readout are adapted from SemIf by TheoLeeCJ (MIT).

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.

JevK5 readme and all models