Assignment #5 - Improving training efficiency
Due Date: 02:00 PM, Thursday, Nov 14
Download [questions] [starter code] [submission template]
In this homework, you will explore strategies for making language models more efficient both in their training and inference phases. Large language models require significant computational resources due to their high memory consumption, large parameter counts, and the extensive compute needed for both training and serving. Efficiently training these models involves optimizing memory usage and using distributed training setups, which allow us to leverage multiple GPUs to process larger models or batches without running out of memory.