Agent Skills: kvpress

kvpress (NVIDIA) KV-cache compression for HuggingFace LLMs. Use when: kvpress imports, compression_ratio, press(model) context managers, StreamingLLMPress, SnapKVPress, ExpectedAttentionPress, TOVAPress, KnormPress, KV-cache eviction, token pruning during generation, or attention sink methods.