3 ms·Accelerating Large Language Model Decoding with Speculative Sampling1 points by georgehill 4y ago