Open fdggdfgg opened 1 year ago
To change cuda, please check the following:
pip install -r requirements/requirements.gpu-cu90.txt
to your versionmodule load cuda90
but you can change it to what's needed on your system. Hope that helps! Note that this repo is a bit old and we've moved on to sockeye version 3 (https://github.com/kevinduh/sockeye-recipes3); it doesn't implement curriculum.sh
but if someone is interested in replicating it for version 3, send in a pull request!
To change cuda, please check the following:
- During installation, modify install/install_sockeye_custom.sh by changing
pip install -r requirements/requirements.gpu-cu90.txt
to your version- During runtime, make sure path to your cuda libraries are available, by adding it to LD_LIBRARY_PATH and/or making it available in scripts/get-device.sh. We use
module load cuda90
but you can change it to what's needed on your system.Hope that helps! Note that this repo is a bit old and we've moved on to sockeye version 3 (https://github.com/kevinduh/sockeye-recipes3); it doesn't implement
curriculum.sh
but if someone is interested in replicating it for version 3, send in a pull request!
Thank you very much for your help
I have another question about how to obtain score files for other parallel corpora.
I tried using the score. sh,but it needs to get model.dir.
I'm not sure how to obtain the model
If Cuda is 11.8, how can I change the version of other packages to make the code run properly. I tried to change the version of another package, but ran../..// When exporting mxnet packages using the scripts/train curriculum.sh - p rs1.hpm - e sockeye curriculum command, an error about cuda may occur