Loading…
Take part
One command runs as many workers as you like; Ctrl-C hands unfinished work back to the queue at once, nothing is lost. Setup (once): a C compiler, Python 3 and git; then
git clone https://github.com/DarthCalculus/Pathology-Puzzle-Searcher.git && cd Pathology-Puzzle-Searcher/backsearch && ./build_pgo.sh -o backsearch_worker_nt --no-torch
Run (and re-run after any git pull; the client refuses a worker build the campaign does not accept, and says how to update):
The client opens a small local page at http://localhost:8765/ with Pause / Resume / Stop and your workers' boards.
No account: your name is just a label on this page. When the campaign is complete the client delivers its last
results and exits by itself (--keep-going waits for the next one). Full guide:
backsearch/v2/README.md,
or the three steps on the collective search page.
Big jobs. A few jobs need more solver memory than an ordinary worker allows; they wait for a
machine with the big worker. One big job can use up to about 5 GB and runs alone (the client's other workers wait,
and it goes back to ordinary jobs when no big job is left). With 8 GB or more to spare, build it once in backsearch
with ./build_pgo.sh --big --no-torch and add --prefer-big to the run command (or pick "Big jobs"
in the client's page).
Per exit
A job is one piece of the search. done: searched to the end; split: handed its unfinished part back as new jobs.
Longest level per root
Each bar is one root job, a starting point of the search listed in the campaign's root file, in listing order. You can hover a root for more information.
Loading…
Roots
Loading…
Explore the search
Pick an exit, then follow moves. Each move's bar shows the CPU time spent below it.
Longest levels so far
Live wall
Leaderboard
CPU time given to this campaign.