espetro / kev-0.8b-gguf

Kev 0.8B — packed GGUF (q8_0)

Kev-0.8B is a Jev-style "System One" decision model: a Qwen3.5-0.8B backbone plus a trained pointer head that returns calibrated probabilities for typed questions instead of generating text.

This GGUF has the pointer head and calibration temperature baked in ( dec.head_* tensors + kev.* metadata), produced by espetro/llama.cpp 's tools/kev/kev_pack.py from the taigrr/kev-0.8b-gguf gojev bundle ( jaredpalmer/kev-0.8b checkpoint).

Use with the Kev-enabled llama.cpp fork

llama-server -hf espetro/kev-0.8b-gguf
# -> serves TypeSafe /v1/systemone + /studio automatically

curl localhost:8080/v1/systemone -H 'content-type: application/json' -d '{
  "state": "Shoes arrived two weeks late and in the wrong size.",
  "questions": {
    "department": {"type": "choice", "instructions": "Which team handles this?",
                   "criteria": {"returns": "Exchanges, refunds", "shipping": "Delays, lost packages"}}
  }
}'

llama-decide -hf espetro/kev-0.8b-gguf --json request.json

The file still loads in stock llama.cpp as an ordinary Qwen3.5 model — the kev.* metadata and head tensors are simply ignored there.

In the browser

espetro.github.io/llama.cpp lazy-loads this file into a llama.cpp WASM build (needs ~1 GB live, q8_0 is the only recommended browser quant — q4 drifts enough to break the calibration).

Source & license

• Weights + head: Apache-2.0 — jaredpalmer/kev
• Source bundle: taigrr/kev-0.8b-gguf (gojev, 0BSD)
• Pack/quantize + runtime: espetro/llama.cpp kev branch (MIT)