Fixstars Corporation, a performance engineering technology company listed on TSE Prime with U.S. headquarters in Irvine, California, announced the release of the newest version of its AI acceleration platform, AIBooster. The update introduces enhanced flexibility and advanced diagnostic capabilities, including improved reports for AcuiRT, AIBooster’s AI model conversion framework optimized for inference, and customizable performance observability with user-defined metrics and tags. These enhancements allow engineers to reduce monitoring overhead and accelerate development cycles by quickly identifying and addressing bottlenecks in model conversion. Fixstars continues to focus on providing tools for efficient, high-performance AI operations that ensure seamless deployment across varied environments.
Health Technology Insights: NeuraLight Unveils Scientific Leaders Advancing Neurology
The latest AIBooster version strengthens diagnostic reporting for AcuiRT, giving engineers the ability to troubleshoot deployment problems through detailed command line output or comprehensive reports. These reports include conversion result visualization with layer-by-layer success rates, error messages for failed layers, and comparisons of inference accuracy to detect potential degradation. Performance profiling is also included, providing overall inference latency and detailed per-layer processing times to help pinpoint bottlenecks. Additionally, performance observability has been made more flexible, allowing users to adjust metric collection intervals to reduce resource consumption and define custom tags for traces. These tags let users classify workloads by model type, execution settings, or dataset, enabling more granular performance analysis.
A practical example of the update’s benefits involved a 2D object detection model, DETR, converted from PyTorch to NVIDIA TensorRT using AcuiRT. Initially, only 16 percent of layers were successfully converted, resulting in lower inference performance. By using the enhanced diagnostic report to identify failed layers and errors, the model was refactored, achieving 100 percent layer conversion in four hours. The optimized model demonstrated a 1.25 times improvement in inference speed, showing the power of the new AIBooster features for faster, more reliable AI model conversion on NVIDIA GPUs.
Health Technology Insights: Rona Therapeutics Advances INHBE siRNA Into Phase 1 Clinical Development
To participate in our interviews, please write to our HealthTech Media Room at info@intentamplify.com
