You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 462dc0f
Browse filesBrowse the repository at this point in the historyBrowse files
feat(unofficial): default to 8K/256 + Output Token Throughput per GPU on unofficial-run load
Temporary branch-only override: when a `?unofficialrun=` URL loads and the
user did not pin `i_seq` / `i_metric` themselves, default the sequence to
`8K / 256` and the y-axis metric to `Output Token Throughput per GPU` so
InfiniteBench-style sweeps land on a useful default view alongside the
model auto-switch from #243.
Mirrors the dedupe shape of `computeAutoSwitchDecision` via a new pure
helper `computeUnofficialOverrideDecision` so manual user picks stick
once URL-synced and a fresh run-set transition can re-arm the override.
Co-authored-by: Bryan Shan <Oseltamivir@users.noreply.github.com>
0 commit comments