Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention
This addresses the challenge of handling extremely long contexts in LLMs for applications like document analysis and streaming inference, representing a novel method rather than an incremental improvement.
The authors tackled the problem of scaling Transformer-based LLMs to infinitely long inputs with bounded memory and computation by introducing Infini-attention, which incorporates compressive memory and combines local and long-term attention mechanisms, achieving results on benchmarks like 1M sequence length retrieval and 500K length summarization with 1B and 8B models.
This work introduces an efficient method to scale Transformer-based Large Language Models (LLMs) to infinitely long inputs with bounded memory and computation. A key component in our proposed approach is a new attention technique dubbed Infini-attention. The Infini-attention incorporates a compressive memory into the vanilla attention mechanism and builds in both masked local attention and long-term linear attention mechanisms in a single Transformer block. We demonstrate the effectiveness of our approach on long-context language modeling benchmarks, 1M sequence length passkey context block retrieval and 500K length book summarization tasks with 1B and 8B LLMs. Our approach introduces minimal bounded memory parameters and enables fast streaming inference for LLMs.