Tag
Trellis introduces RadixAttention to optimize LLM inference prefill phase by caching prefix tokens using a radix tree, reducing redundant computation in chat and agentic sessions.