![]() ![]() Everyone reports that Monterey has considerable performance improvements over Big Sur, so at this point I am going to skip Big Sur altogether. You can take a look at the DLC requirements here: DeepLabCut/DeepLabCut/blob/1c9f59818ba2e960827167f52715e127eebd63bd/requirements. Personally what I would do (which is exactly what I am doing) is to stay on Catalina until Monterey has fixed some of its current issues. The idea there is to recreate what DLC needs in a native arm64 python 3.8 (or 3.9 – I’ve been using 3.9 exclusively for tensorflow) env with Apple tensorflow and the metal plugin. 08:16:23.926585: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. ![]() I have only run testscript.py and not a full project, and it only runs in “light mode” (no GUI) but, the GPU was engaged during the training sets (python GPU usage went to 95%) and the following confirmation statements. I think I got DLC to install on an M1 macbook pro, running macOS Monterey 12.0.1. Installing DLC on mac with M1 and engaging GPU Development Python -m pip install tensorpack tf_slim tables conda-environments/DEEPLABCUT_M1.yamlĬonda install filterpy ruamel.yaml imgaug numba scikit-image scikit-learn statsmodels tqdm moviepy We are currently working with a local copy of dlc (but identical to the main GITversion) and the following code successfully installed everything (including the GUI): conda env create -f. My jobs are running ok on colab, but I really really prefer to run in an environment I can control. When the training ends the memory is not released and the only way I recover it is by ctrl-Z the pythonw. The same task now is using 50GB and running at 20-30% when I’m lucky. Unfortunately I did not record enough details on Big Sur, but my recollection is I was using most of the 16GB I have on this M1 mini, and the GPU was running at above 90%. ![]() Open the apps that you want to use with the eGPU. Select the display thats attached to the eGPU, then choose Use as Main Display. Choose Apple menu > System Settings (or System Preferences), then click Displays. A training that took ~10 minutes before is taking 6 hours now. In macOS Monterey and later: Quit any open apps that you want the eGPU to accelerate on the primary display. Once we switched to Monterey it is a disaster. I have succeeded in running dlc on Big Sur, very efficiently. ![]()
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |