Existing methods for speeding up pre-filling often fail to maintain acceptable accuracy or efficiency when applied to long-context LLMs.

The computational challenges of Large Language Model (LLM) inference remain a significant barrier to their widespread deployment, especially as prompt lengths continue to increase. Due to the quadratic complexity of the attention computation, it takes 30 minutes for an 8B LLM to process a prompt of 1M tokens (i.e., the pre-filling stage) on a single A100 GPU. Existing methods for speeding up prefilling often fail to maintain acceptable accuracy or efficiency when applied to long-context LLMs. To address this gap, we introduce MInference (Milliontokens Inference), a sparse calculation method designed to accelerate pre-filling of long-sequence processing. Specifically, we identify three unique patterns in long-context attention matrices-the A-shape, Vertical-Slash, and Block-Sparse-that can be leveraged for efficient sparse computation on GPUs.

Memory, additional information beyond the training of large language models (LLMs), is crucial to various real-world applications, such as personal assistant. The two mainstream solutions to incorporate memory into the generation process are long-context LLMs and retrieval-augmented generation (RAG). In this paper, we first systematically compare these two types of solutions on three renovated/new datasets and show that (1) long-context solutions, although more expensive, shall be easier to capture the big picture and better answer queries which require considering the memory as a whole; and (2) when the queries concern specific information, RAG solutions shall be more competitive especially when the keywords can be explicitly matched. Therefore, we propose a novel method RAG-Tuned-LLM which fine-tunes a relative small (e.g., 7B) LLM using the data generated following the RAG principles, so it can combine the advantages of both solutions. Extensive experiments on three datasets demonstrate that RAG-Tuned-LLM can beat long-context LLMs and RAG methods across a wide range of query types.

The expansion of large language models (LLMs) with hundreds of billions of parameters presents significant challenges to computational resources, particularly data movement and memory bandwidth. Long-context LLMs, which process sequences of tens of thousands of tokens, further increase the demand on the memory system as the complexity in attention layers and key-value cache sizes is proportional to the context length. Processing-in-Memory (PIM) maximizes memory bandwidth by moving compute to the data and can address the memory bandwidth challenges; however, PIM is not necessarily scalable to accelerate long-context LLM because of limited per-module memory capacity and the inflexibility of fixed-functional unit PIM architecture and static memory management. In this work, we propose LoL-PIM which is a multi-node PIM architecture that accelerates long context LLM through hardware-software co-design. In particular, we propose how pipeline parallelism can be exploited across a multi-PIM module while a direct PIM access (DPA) controller (or DMA for PIM) is proposed that enables dynamic PIM memory management and results in efficient PIM utilization across a diverse range of context length. We developed an MLIR-based compiler for LoL-PIM extending a commercial PIM-based compiler where the software modifications were implemented and evaluated, while the hardware changes were modeled in the simulator. Our evaluations demonstrate that LoL-PIM significantly improves throughput and reduces latency for long-context LLM inference, outperforming both multi-GPU and GPU-PIM systems (up to 8.54x and 16.0x speedup, respectively), thereby enabling more efficient deployment of LLMs in real-world applications.

We introduce a bootstrapping approach to train long-context language models by exploiting their short-context capabilities only. Our method utilizes a simple agent workflow to synthesize diverse long-context instruction tuning data, thereby eliminating the necessity for manual data collection and annotation. The proposed data synthesis workflow requires only a short-context language model, a text retriever, and a document collection, all of which are readily accessible within the open-source ecosystem. Subsequently, language models are fine-tuned using the synthesized data to extend their context lengths. In this manner, we effectively transfer the short-context capabilities of language models to long-context scenarios through a bootstrapping process. We conduct experiments with the open-source Llama-3 family of models and demonstrate that our method can successfully extend the context length to up to 1M tokens, achieving superior performance across various benchmarks.

Retrieval-augmented generation (RAG) has gained traction as a powerful approach for enhancing language models by integrating external knowledge sources. However, RAG introduces challenges such as retrieval latency, potential errors in document selection, and increased system complexity. With the advent of large language models (LLMs) featuring significantly extended context windows, this paper proposes an alternative paradigm, cache-augmented generation (CAG) that bypasses real-time retrieval. Our method involves preloading all relevant resources, especially when the documents or knowledge for retrieval are of a limited and manageable size, into the LLM's extended context and caching its runtime parameters. During inference, the model utilizes these preloaded parameters to answer queries without additional retrieval steps. Comparative analyses reveal that CAG eliminates retrieval latency and minimizes retrieval errors while maintaining context relevance. Performance evaluations across multiple benchmarks highlight scenarios where long-context LLMs either outperform or complement traditional RAG pipelines. These findings suggest that, for certain applications, particularly those with a constrained knowledge base, CAG provide a streamlined and efficient alternative to RAG, achieving comparable or superior results with reduced complexity.

Large Language Models (LLMs) have shown exciting performance in listwise passage ranking. Due to the limited input length, existing methods often adopt the sliding window strategy. Such a strategy, though effective, is inefficient as it involves repetitive and serialized processing, which usually re-evaluates relevant passages multiple times. As a result, it incurs redundant API costs, which are proportional to the number of inference tokens. The development of long-context LLMs enables the full ranking of all passages within a single inference, avoiding redundant API costs. In this paper, we conduct a comprehensive study of long-context LLMs for ranking tasks in terms of efficiency and effectiveness. Surprisingly, our experiments reveal that full ranking with long-context LLMs can deliver superior performance in the supervised fine-tuning setting with a huge efficiency improvement. Furthermore, we identify two limitations of fine-tuning the full ranking model based on existing methods: (1) sliding window strategy fails to produce a full ranking list as a training label, and (2) the language modeling loss cannot emphasize top-ranked passage IDs in the label. To alleviate these issues, we propose a new complete listwise label construction approach and a novel importance-aware learning objective for full ranking. Experiments show the superior performance of our method over baselines. Our codes are available at \url{https://github.com/8421BCD/fullrank}.

Long-context large language models (LC LLMs) promise to increase reliability of LLMs in real-world tasks requiring processing and understanding of long input documents. However, this ability of LC LLMs to reliably utilize their growing context windows remains under investigation. In this work, we evaluate the performance of state-of-the-art GPT-4 suite of LC LLMs in solving a series of progressively challenging tasks, as a function of factors such as context length, task difficulty, and position of key information by creating a real world financial news dataset. Our findings indicate that LC LLMs exhibit brittleness at longer context lengths even for simple tasks, with performance deteriorating sharply as task complexity increases. At longer context lengths, these state-of-the-art models experience catastrophic failures in instruction following resulting in degenerate outputs. Our prompt ablations also reveal unfortunate continued sensitivity to both the placement of the task instruction in the context window as well as minor markdown formatting. Finally, we advocate for more rigorous evaluation of LC LLMs by employing holistic metrics such as F1 (rather than recall) and reporting confidence intervals, thereby ensuring robust and conclusive findings.

The development of Long-Context Large Language Models (LLMs) has markedly advanced natural language processing by facilitating the process of textual data across long documents and multiple corpora. However, Long-Context LLMs still face two critical challenges: The lost in the middle phenomenon, where crucial middle-context information is likely to be missed, and the distraction issue that the models lose focus due to overly extended contexts. To address these challenges, we propose the Context Filtering Language Model (FltLM), a novel integrated Long-Context LLM which enhances the ability of the model on multi-document question-answering (QA) tasks. Specifically, FltLM innovatively incorporates a context filter with a soft mask mechanism, identifying and dynamically excluding irrelevant content to concentrate on pertinent information for better comprehension and reasoning. Our approach not only mitigates these two challenges, but also enables the model to operate conveniently in a single forward pass. Experimental results demonstrate that FltLM significantly outperforms supervised fine-tuning and retrieval-based methods in complex QA scenarios, suggesting a promising solution for more accurate and reliable long-context natural language understanding applications.