-
Woops
thomasw21
-
Victor suggestion to send the last batch
thomasw21
-
Get PMD running
thomasw21
-
Debug
thomasw21
-
I'm stupid
thomasw21
-
More debug
thomasw21
-
Woops
thomasw21
-
debug
thomasw21
-
remove default logs
thomasw21
-
Woops
thomasw21
-
Run everything in parallel except sqlite3
thomasw21
-
Logs
thomasw21
-
Woops
thomasw21
-
Woops
thomasw21
-
Woops
thomasw21
-
Test out arrowBuilder
thomasw21
-
Debug proc
thomasw21
-
Debug low throughput
thomasw21
-
Maybe I need to load everything into memory first
thomasw21
-
Use a lot bigger batch size
thomasw21
-
Woops
thomasw21
-
Perhaps that's better
thomasw21
-
Define a batch_size_per_proc to lower memory footprint, there's going to be a natural tradeoff
thomasw21
-
Reduce chunk size
thomasw21
-
Maybe fix
thomasw21
-
Add support for multiprocessing for visual genome
thomasw21
-
Add back all configs
thomasw21
-
Remove prints
thomasw21
-
missing
thomasw21
-
Debug the filtering of yfcc100m
thomasw21
-
Test yfcc100m
thomasw21
-
Being able to seriablize
thomasw21
-
Debug conceptual_12m
thomasw21
-
PMD dataset
thomasw21
-
initial commit
Thomas Wang