Fetching the paper…

KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization · Around