
Recent advancements in long-context language models (LLMs) have been highlighted, focusing on improving their ability to handle extended contexts. MemLong, a memory-augmented retrieval system, enhances LLMs by using an external retriever and a memory bank, significantly extending context length from 4,000 to 80,000 tokens. This development allows LLMs to outperform other state-of-the-art models on long-context benchmarks, even on a single 3090 GPU. Additionally, new methods such as Writing in the Margins, ReMamba, Dolphin, FocusLLM, and LongRAG have been introduced to improve the efficiency of processing long contexts in LLMs.
MemLong utilizes an external retriever for retrieving historical information which enhances the capabilities of long-context LLMs. It consistently outperforms other SoTA LLMs on long-context benchmarks and can extend the context length on a single 3090 GPU from 4k up to 80k. https://t.co/FmeSv9tMSB
Long-context LLMs Struggle with Long In-context Learning https://t.co/VRKdECKOIu #longiclbench #learning #long
How a research scientist at Google DeepMind use LLMs Really cool read about the practical things LLMs do rly well: https://t.co/vVBnUm5xXM https://t.co/7u5J3cJlwB



