CentML provides a platform and compiler that identifies training bottlenecks, predicts total time and cost to deploy models, and automatically optimizes model training workloads to run better on target GPUs. The company says its optimizations can cut expenses by up to 80% without degrading model accuracy and cites a customer case that ran Llama 2 three times faster on Nvidia A10 cards. CentML’s compiler produces optimized code for a variety of GPUs and aims to reduce memory needed to deploy models so teams can use smaller, cheaper hardware. The startup competes with firms like MosaicML and OctoML and claims its techniques avoid accuracy loss and offer a newer-generation compiler. In the near term CentML plans to expand its work from training optimizations into inference optimization. The company has a workforce of about 30 people across the U.S. and Canada and will use new capital to bolster product development, research, and engineering headcount.

Total raised

$27M

Funding rounds

1

Latest round

Seed

Latest activity

Oct 2023