N
Hacker Next
new
past
show
ask
show
jobs
submit
login
▲
Codex reasoning-token clustering at 516 may be leading to degraded performance
(
github.com
)
11 points by
0x_rs
6 hours ago
|
1 comment
add comment
Rendered at 23:13:29 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
cyanydeez 2 hours ago
[-]
wonder if theyre basically doing this llamacpp reasoning trick:
https://github.com/ggml-org/llama.cpp/blob/master/common/rea...
id guess its the harness
id guess its the harness