Open bvillasen opened 11 months ago
Sorry for the delay. Just a reminder that some legitimate code changes do change the expected values of the benchmark results (especially for the tiny case, since there are so few input samples). We do try to update the bundled "expected" values when making a tag, but not for every commit (since it requires running a bunch of jobs that do not fit within CI runners). The benchmark scripts are bundled with the installed package, and you should only use the included versions of those scripts. In other words, after installing a new version of toast, re-run the setup and benchmark scripts that are installed with that version.
Having said that, I did not update the bundled "expected" values before the last tag, and there were some changes to those expected values. I'll make a new tag in the next day or so with updated values. I'll ping this issue when it is in place.
I'm building and running the toast3 branch of this repository following the instructions presented HERE
When I build and run toast_benchmark_ground_setup from the upstream version of the toast3 branch for the tiny workload, I get the following output:
The solver converges after seven iterations, and the resulting statistics don't match the expected values. Same result is obtained when building and running the version corresponding to commit 24431d12f82ae827176f47af4d13506988948127 (Oct 2023)
When I build and run toast_benchmark_ground_setup from an older version corresponding to commit 0992fe001cf20378204d835e9fa99b36c99ec181 (Feb 2023), the output is the following:
Here, the solver finishes in 30 iterations, and the resulting statistics match the expected values.
I also note that the wall time per iteration is less than half in the current version compared to the older commit.
Can you please confirm that the current version of toast3 is running correctly?