Patent Leaderboard
USPTO Patent Rankings Data through Dec 31, 2025
Home › Scaling half-precision floating point tensors for training deep neural networks

Scaling half-precision floating point tensors for training deep neural networks

US Patent 12106210 · Granted Oct 1, 2024

Estimated economic value: $20,560,000 ⓘ

Assignee

Inventors

View full patent text on Google Patents →

More Patents by These Inventors

System and method for automated calibration for a scanning system2025
Communication optimizations for distributed machine learning2025
Instructions to convert from FP16 to BF82025
Instructions for structured-sparse tile matrix FMA2025
Dynamic precision management for integer deep learning primitives2025
Dynamic precision management for integer deep learning primitives2025
Utilizing structured sparsity in systolic arrays2025
Utilizing structured sparsity in systolic arrays2025
Systolic array having support for output sparsity2025
Systolic array having support for output sparsity2025
Abstraction layers for scalable distributed machine learning2025
Instructions to convert from FP16 to BF82025