Using CoreWeave Mission Control to Diagnose and Fix Slow Inference
CoreWeave Mission Control AI diagnoses slow inference by identifying undersized KV cache capacity. The fix involves quantizing model weights from FP8 to NVFP4, nearly doubling throughput to 18,400 tokens/second.
Read article