Made a very stupid sampler-based lever to try and mimic o1-style "reasoning_effort=high" on r1: If </think> appears before enough thinking tokens have been generated, the sampler replaces it…

Posters: voooooogel. voooooogel

3 likes recorded by the source corpus.

Topics: AI & LLMs

View Post: https://x.com/voooooogel/status/1881966969043464365