A Software First Approach
A complete software stack for SambaNova DataScale, SambaFlow™ is designed to advance developer productivity. Built to fully integrate with popular standard frameworks such as PyTorch and TensorFlow, SambaFlow provides an open, flexible, and easy-to-use development interface for exploring the new capabilities unlocked by DataScale.
SambaFlow automatically extracts, optimises, and executes the optimal dataflow graph of any of your models on SambaNova’s Reconfigurable Dataflow Units (RDUs). This enables you to achieve out-of-the-box performance, accuracy, scale, and ease of use. With SambaFlow, you can maximise productivity by focusing on your development in the frameworks without ever again worrying about low-level tuning.
Ease of Use
SambaFlow manages the DataScale platform from frameworks to hardware
- Fully integrated with popular open source ML frameworks, such as
PyTorch and TensorFlow. No code required to run your existing models.
- Automated data and model parallel mapping simplifies scaling by using
the same programming model you would use for a single device across
any number of devices and configurations.
- High-level API provides power users with additional control without
needing to understand the details of hardware implementation.
- Native integration with developer productivity repos, such as Hugging
Face. Run thousands of models at state-of-the-art performance and
accuracy in seconds with zero code change.