ggml-org/whisper.cpp
Should we use process time for benchmarks/timings?
Offen
#382 geöffnet am 07.01.2023
enhancementgood first issue
Repository-Metriken
- Stars
- (52.942 Sterne)
- PR-Merge-Metriken
- (Durchschn. Merge 5T 20h) (36 gemergte PRs in 30 T)
Beschreibung
Currently, the bench tool simply spits out the timings from whisper_print_timings. These timings are not process specific and as such are influenced by other processes on the system.
Should we use something like CLOCK_PROCESS_CPUTIME_ID instead of CLOCK_MONOTONIC in ggml_time_*? This would make the benchmarks in #89 outdated, but would make future benchmarks more realistic.