{"id":1923,"date":"2021-02-01T14:37:52","date_gmt":"2021-02-01T13:37:52","guid":{"rendered":"https:\/\/eurocc.nscc.sk\/?page_id=1923"},"modified":"2021-02-01T14:45:06","modified_gmt":"2021-02-01T13:45:06","slug":"horovod","status":"publish","type":"page","link":"https:\/\/eurocc.nscc.sk\/en\/services\/manualy\/horovod\/","title":{"rendered":"Horovod"},"content":{"rendered":"<div class=\"is-layout-flow wp-block-group manuals\"><div class=\"wp-block-group__inner-container\">\n<h4>Horovod<hr style=\"border: 1px solid #b8870b; width: 100px;\"><\/h4>\n\n\n\n<p>Pokia\u013e potrebujete pou\u017ei\u0165 machine learning model cez viacero n\u00f3dov a vyu\u017ei\u0165 tak viacero GPU, jednou z mo\u017enost\u00ed je <a href=\"https:\/\/horovod.ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">horovod<\/a>. <\/p>\n\n\n\n<p>Horovod je open source softv\u00e9r pre distribuovan\u00fd deep learning tr\u00e9ning pomocou Tensorflow, Keras, PyTorch a Apache MXNET. Jeho cie\u013eom je zlep\u0161enie r\u00fdchlosti, \u0161k\u00e1lovate\u013enosti a alok\u00e1cie zdrojov pre potreby tr\u00e9novania machine learning modelov.<\/p>\n\n\n\n<p><strong>Pou\u017eitie<\/strong><\/p>\n\n\n\n<p>Ak chcete pou\u017ei\u0165 horovod na v\u00fdpo\u010dtovom klastri v Ko\u0161iciach, je potrebn\u00e9 si najprv upravi\u0165 v\u00e1\u0161 python zdrojov\u00fd k\u00f3d pod\u013ea in\u0161trukci\u00ed na https:\/\/github.com\/horovod\/horovod#usage. Na ko\u0161ickom klastri je Horovod nain\u0161talovan\u00fd s podporou OpenMPI, sp\u00fa\u0161\u0165a sa podobne ako ostatn\u00e9 MPI aplik\u00e1cie. Horovod sa automaticky sna\u017e\u00ed spusti\u0165 jeden viacvl\u00e1knov\u00fd proces na v\u00fdpo\u010dtov\u00fd n\u00f3d.<\/p>\n\n\n\n<p>Pr\u00edklad pre tensorflow n\u00e1jdete v \/lustre\/home\/freeware\/EXAMPLE_JOBS\/horovod. Sp\u00fa\u0161\u0165ac\u00ed skript pre slurm je run.sh. Vy\u017eiada dva n\u00f3dy, celkovo dva procesy a jeden proces na n\u00f3d, po\u017eiada iba o n\u00f3dy s grafickou kartou k20m:<\/p>\n\n\n\n<p>#\u2014-run.sh\u2014-<\/p>\n\n\n\n<p>#!\/bin\/bash<\/p>\n\n\n\n<p>#SBATCH \u2013partition=short<\/p>\n\n\n\n<p>#SBATCH \u2013job-name=test<\/p>\n\n\n\n<p>#SBATCH \u2013output=out.txt<\/p>\n\n\n\n<p>#SBATCH \u2013error=err.txt<\/p>\n\n\n\n<p>#SBATCH \u2013nodes=2<\/p>\n\n\n\n<p>#SBATCH \u2013ntasks=2<\/p>\n\n\n\n<p>#SBATCH \u2013ntasks-per-node=1<\/p>\n\n\n\n<p>#SBATCH \u2013time=30:00<\/p>\n\n\n\n<p>#SBATCH \u2013constraint=k20mmpirun python3 test_keras.py<\/p>\n\n\n\n<p>Horovod je mo\u017en\u00e9 okrem OpenMPI pou\u017ei\u0165 aj s mpi4py alebo singularity kontajnermi. Kompletn\u00fa dokument\u00e1ciu n\u00e1jdete na <a href=\"https:\/\/github.com\/horovod\/horovod\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/horovod\/horovod<\/a>.<\/p>\n<\/div><\/div>","protected":false},"excerpt":{"rendered":"<p>Horovod Pokia\u013e potrebujete pou\u017ei\u0165 machine learning model cez viacero n\u00f3dov a vyu\u017ei\u0165 tak viacero GPU, jednou z mo\u017enost\u00ed je horovod. Horovod je open source softv\u00e9r pre distribuovan\u00fd deep learning tr\u00e9ning pomocou Tensorflow, Keras, PyTorch a Apache MXNET. Jeho cie\u013eom je zlep\u0161enie r\u00fdchlosti, \u0161k\u00e1lovate\u013enosti a alok\u00e1cie zdrojov pre potreby tr\u00e9novania machine learning modelov. Pou\u017eitie Ak chcete [&hellip;]<\/p>","protected":false},"author":1,"featured_media":0,"parent":1524,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"templates\/template-full-width.php","meta":[],"_links":{"self":[{"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/pages\/1923"}],"collection":[{"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/comments?post=1923"}],"version-history":[{"count":3,"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/pages\/1923\/revisions"}],"predecessor-version":[{"id":1933,"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/pages\/1923\/revisions\/1933"}],"up":[{"embeddable":true,"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/pages\/1524"}],"wp:attachment":[{"href":"https:\/\/eurocc.nscc.sk\/en\/wp-json\/wp\/v2\/media?parent=1923"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}