Does Your Model Run on AWS Trainium? Look It Up
huggingface ·
Hugging Face has introduced a new Space to help users determine if their models run on AWS Trainium using the TorchNeuron backend. This tool allows users to look up popular models by name or browse trending and downloaded lists to see compatibility status. The interface displays key metrics such as the number of Trainium cores used, compilation time, and peak memory usage for each tested model. Users can also access the specific Python scripts required to run these models themselves on Trainium2 instances. The testing process involves loading models with tensor or expert parallelism and verifying output accuracy against CPU runs or fluency checks. Currently, the database focuses on text-generation and image-text-to-text models that fit within four Trainium2 devices. All test results are stored in a public dataset, enabling developers to query compatibility data directly via code.
huggingface