What these adapters were measured against

#1
by YauhenBichel - opened

These adapters are the 0.5B style prior for python-vibe. They are not the daily model.

Same project, other weights:

  • llama3.1:8b β€” daily default. Same-night jobs 9/9.
  • qwen2.5-coder:7b β€” comparison. 7/9 that night. Do not switch.
  • Qwen2.5-Coder-32B-Instruct β€” hosted. 1/10 then 9/10 after markdown fences were stripped. The local 8B does not fence, so that fix does not move it.

This repo: 0/4 vibe, 0/54 greedy LoRA.

Code: https://github.com/YauhenBichel/python-vibe

The GitHub repo is now py-harness (formerly python-vibe).

https://github.com/YauhenBichel/py-harness
https://yauhenbichel.github.io/py-harness/

Hub adapter id is unchanged: YauhenBichel/python-vibe-0.5b

Sign up or log in to comment