Prostt5 ctranslate2 splitting long sequenes#327

Open

mpjw wants to merge 23 commits intosteineggerlab:prostt5-ctranslate2from mpjw:prostt5-ctranslate2

Commits on Aug 6, 2024

developed draft workflow for splitting sequences bevore prostt5 prediction
Marcus Wagner
committed

Commits on Aug 9, 2024

splitting of long sequences compiles and runs, but there is a sequence length bug in the 3Di predictions
Marcus Wagner
committed
fixed length mismatch of predicted 3Di seqs
Marcus Wagner
committed
cleaning up comments from debugging
Marcus Wagner
committed
testing README and data
Marcus Wagner
committed

Commits on Aug 12, 2024

added splitting parameter with max 6000 lenth
Marcus Wagner
committed
removing limit of 6000 for split parameter
Marcus Wagner
committed
creating minimal conda env for CUDA compiling
Marcus Wagner
committed

Commits on Aug 13, 2024

renamed cuda developement environment
Marcus Wagner
committed
renamed long AA fasta, added second split test for 6000 length splits
Marcus Wagner
committed
creating minimal conda env
Marcus Wagner
committed
disableing thread affinity due to std::runtime_error thrown when testing splitting functionality
Marcus Wagner
committed
gitignore file for build, test, and other dev folders
Marcus Wagner
committed
polishing README
Marcus Wagner
committed

Commits on Aug 14, 2024

updating test data for 6000 split length case
Marcus Wagner
committed
split specific length calculation for substring creation
Marcus Wagner
committed
eliminating MAX_SPLIT_LENGTH constant, by using the parameter multiple times
Marcus Wagner
committed
final working tests
Marcus Wagner
committed

Commits on Aug 21, 2024

foldseek paper benchmark for ctranslate2 splitting branch
Marcus Wagner
committed

Commits on Aug 22, 2024

enabling CUDA for ProstT5
Marcus Wagner
committed
working on split length benchmark for foldseek ctranslate2
Marcus Wagner
committed
foldseek crashed with CUDA enabled for ProstT5; currently debugging
Marcus Wagner
committed
rocx files for benchmarking; these are wrongly scaled, need to recompute once CUDA build is fixed
Marcus Wagner
committed