search

google / deepconsensus

DeepConsensus uses gap-aware sequence transformers to correct errors in Pacific Biosciences (PacBio) Circular Consensus Sequencing (CCS) data.
BSD 3-Clause "New" or "Revised" License
222 stars 37 forks source link

vRAM limit #64

Closed fergsc closed 1 year ago

fergsc commented 1 year ago

Hi,

I am trying to understand what the comutational limits for deepconsensus are. Does the batch size of ZMW --batch_zmws=100 control how much data is passed to the GPU? Ie. if we have a lot of RAM on GPU we can increase this?

Thanks.

akolesnikov commented 1 year ago

Hi @fergsc,

There are two batch parameters: --batch_zmws and --batch_size. --batch_size controls how many examples are processed in one call to the TensorFlow model. You may try that to increase the load on GPU.

--batch_zmws controls how many ZMWs are processed at a time. For each ZMW batch preprocess and inference steps are called sequentially. --cpus flag controls how many processes are used for preprocessing step. This flag does not affect the inference (run on GPU) part.