#MatrixOperations
Deep dive into trainable self-attention: how LLMs process token relationships through matrix operations
https://www.gilesthomas.com/2025/03/llm-from-scratch-8-trainable-self-attention
#machinelearning #neuralnetworks #llmarchitecture #matrixoperations #self-attention
March 5, 2025 at 7:53 AM
DeepGEMM: Fast FP8 matrix multiplication library with runtime compilation and NVIDIA Hopper support
https://github.com/deepseek-ai/DeepGEMM
#cuda #gpucomputing #matrixoperations #performanceoptimization #deeplearning
February 26, 2025 at 2:56 AM
Beyond new algorithms, alternative strategies were mentioned. Sometimes, avoiding explicit operations like matrix inversion or transposition by optimizing data reading/iteration patterns can offer performance gains without changing the core math. #MatrixOperations 7/7
May 17, 2025 at 7:00 PM