3 ms·Real-time LLM Inference on Standard GPUs (3k tokens/s per request)7 points by morgangiraud 4mo ago