On April 23, 2026, a company announced GPU kernels for language-model operations, developed with TileLang. The company says some are already used internally for training and inference.
On April 23, 2026, a company announced Tile Kernels, GPU kernels for language-model operations developed with TileLang. The company says some are used in its internal training and inference processes.
The announcement identifies optimizations engineers can evaluate for language-model training and inference, but the available material provides no performance metrics or comparisons. Consult the original publication to verify its scope and technical details before drawing conclusions.