kermt-finetune

To finetune a protein encoder on a custom CSV dataset, you need a labeled molecular dataset and containerized GPU environment. This skill automates data validation, scaffold-balanced splitting, and multi-GPU training orchestration for the finetuning process.

Yes, multi-GPU training for molecular property prediction is supported via Docker. The workflow utilizes multi-GPU DDP support and requires the NVIDIA Container Toolkit alongside CUDA-capable hardware to execute the training scripts.

Finetuning KERMT models requires user-provided labeled CSV datasets containing molecular data such as small molecules with binding affinity labels. The workflow includes automated data validation and scaffold-balanced splitting for these CSV files.

Yes, finetuning KERMT encoders supports regression, classification, and multiclass tasks. The workflow provides automated hyperparameter management to adapt the pretrained model to your specific labeled life science dataset.

Yes, CUDA-capable hardware is strictly required to execute kermt-finetune training workflows. You must also install Docker and the NVIDIA Container Toolkit to manage the containerized training environments.

The best way to adapt a pretrained encoder for custom life science tasks is using an automated finetuning workflow. It handles checkpoint validation, data cleaning, feature preparation, and detached training execution without manual script management.