pLM-BLAST
pLM-BLAST copied to clipboard
Multi-GPU support in embeddings.py
python embeddings.py start data_set_fullseq.fasta data_set_fullseq -embedder pt --gpu -bs 0
When run as above, everything works fine. However, if I use the -nproc 2
flag, the computations are much faster, but the resulting database file contains embeddings for only half of the input sequences.
-nproc
will only work correctly when asdir
or h5py
option is specified, this should be validated before actual run
I've stumbled upon this too - there are also 2 overlapping progress bars shown as I recall. Additionally there's a typo in README that I've followed at that time: https://github.com/labstructbioinf/pLM-BLAST/blob/main/README.md?plain=1#L64
Yes, the progress bars looked weird (one was updating during the calculations and the other was stuck at 20%).