4 ms·Messing around with fine-tuning LLMs, detailed memory usage for an 8B model1 points by vednig 2y ago