3 ms·Low-Latency Inference with Speculative Decoding on D-Matrix Corsair and GPU1 points by nserrino 7mo ago