This repository contains the source code for the SmartFESR teachnology stack, a modular and scalable toolset designed to support machine learning applications. The infrastructure is designed to be highly customizable and to support a wide range of use cases.
SmartFESR can be used with Docker. The Docker documentation can be found in the /docker folder.
This repository contains the source code for a replication study of the GenCast model, a state-of-the-art model for weather forecasting. The study uses the SmartFESR toolkit to download and process the data, and to perform the inference using the pre-trained model.
According to the official documentation, the GenCast model in inference with GPU requires at least 300GB of RAM and 60GB of vRAM.
The steps to reproduce the study are as follows:
-
The first step is to download and process the data using the ccai-gencast-data_fusion-app.
-
The second step is to run the pre-trained GenCast model on the processed data using the ccai-gencast-ml-app.
-
The third step is to make predictions accessible using the ccai_gencast-serving-app.
-
The fourth and final step is to visualize the results using the ccai_gencast_frontend-app.
In order to move data from Surface and make them accessible via FESR applications follow these steps:
-
Install and run Surface.
-
Run FTP server.
-
Run an FTP file ingestion with Surface, following this tutorial.
-
Optionally, you can also execute a backup of all Surface data, following this tutorial.
-
Proceed on our Docker documentation.
-
Then, all Surface data can be reached from FESR applications thanks to the ccai_surface-data_fusion-app. In order to do that, watch the last tutorial.
- Developed by Smart Shaped srl
- Funded by REGIONE ABRUZZO A VALERE SUL PR FESR ABRUZZO 2021-2027, CODICE CUP C29J24000080007