Codex reasoning-token clustering at 516 may be leading to degraded performance
1–3 of 3 posts
Re: Codex reasoning-token clustering at 516 may be leading to degraded performance
#2wonder if theyre basically doing this llamacpp reasoning trick: https://github.com/ggml-org/llama.cpp/blob/master/common/rea...
id guess its the harness
Re: Codex reasoning-token clustering at 516 may be leading to degraded performance
#3this is happening on xhigh config!! extremely pissing off