RDX for developers: the dataflow stack behind fast inference

A half-day, hands-on workshop on RDX (Reconfigurable Dataflow eXecution), SambaNova's developer stack for building and optimising models on its Reconfigurable Dataflow Units (RDUs).

You will compile a PyTorch model for SambaNova RDUs with RDX Torch, simulate it with RDX Analyzer and read the utilisation report before it touches hardware. You will then use RDX Partitioner to tile and distribute a model across devices with a DTensor-style workflow, exploring how the order of directives changes the result and affects performance.

Who should attend

• ML and AI engineers
• Compiler and performance engineers
• Platform teams who deploy models for inference

You should be comfortable with Python and PyTorch. Experience with distributed inference will help with the advanced exercises but is not required. Bring a laptop.

About the presenters

Justin Woo

Justin Woo is a Developer Advocate at SambaNova and is passionate about helping developers build and optimise models on SambaNova's Reconfigurable Dataflow Units. He graduated from Purdue University with a degree in Computer Engineering and previously worked at Microsoft and PayPal. He lives with his wife and two dogs—one of which refuses to admit that he is blind and the other that likes to pretend she's a rabbit. In his spare time he loves playing squash, pickleball and board games, with Settlers of Catan a favourite.

MK Swaminathan

MK Swaminathan is Product Lead at SambaNova, where she works on AI inference infrastructure for enterprises and sovereign AI providers. She focuses on helping organisations run AI faster and at lower cost while keeping control of their own data.