KV Cache Quantization

Converting Key and Value tensors in LLM inference to low-precision formats (e.g., FP8) to reduce memory footprint and accelerate