Execution framework for multi-task model parallelism. Enables the training of arbitrarily large models with a single GPU, with linear speedups for multi-gpu multi-task execution.
Primary LanguagePythonApache License 2.0Apache-2.0