2 ms·DeepSpeed-FastGen: High-Throughput for LLMs via MII and DeepSpeed-Inference2 points by CharlesW 3y ago