An analytical performance model and open-source tool for high-level co-design of LLM training and inference systems.
An analytical performance model and open-source tool for high-level co-design of LLM training and inference systems.
We present Calculon, a parameterized analytical performance model and open-source tool for high-level algorithm-architecture codesign of LLM training and inference systems.
We use an analytical performance model to design systems that train LLMs of up to 128 trillion parameters at 75%+ MFU.