loss improvement

optimizer step (1) flops reduction (1) training acceleration (1) low-rank adaptation (1) parameter-efficient fine-tuning (1) language model (1) neural network training (1) stochastic gradient descent (1)