System Requirements

System requirements strongly depend on (1) the domain (text, document, images etc.), (2) whether a model should be trained or new data should be predicted and (3) the type of the model that is used for training. Nevertheless, we experienced in the past that there are standard setups that work pretty well, therefore, we suggest the following setup:

Training

General

  • Python 3.10, 3.11

  • CPU with more than 8 cores

  • System Memory with at least 16 GB

  • At least one Nvidia GPU with at least 8 GB of memory (e.t. 3070, 3080)

DocumentModels

  • We recommend using a GPU with at least 24 GB of memory. Using our model improvements as explained in the installation guide you can also use a GPU with less memory.

  • (Optional) CUDA > 11.4 as well as PyTorch > 1.12 to enable several performance, memory and speedup improvements. See also our installation guide for more details.

Other recommendations

  • We recommend a Linux-based OS

  • SSD with at least 256 GB

Prediction

System requirements strongly depend on the number of predictions that should be executed per second. For several samples each second, we recommend using the same setup as used for training. Otherwise, if predictions are executed only a few times each minute, no GPU is required as transferring data from the GPU to the CPU is larger than the speedup that you gain from the GPU.

In case you are uncertain about what setup to use, feel free to contact us (hello@deepopinion.ai). We are always to discuss and help you with your current task.