Transformer-based Large Language Models (LLMs) are typically pre-trained with a fixed context window size, e.g ., 4K tokens in Touvron et al. [2023a].

Transformer-based Large Language Models (LLMs) are typically pre-trained with a fixed context window size, e.g ., 4K tokens in Touvron et al. [2023a].

Large Language Models (LLMs) often struggle to use information across long inputs effectively. Prior work has identified positional biases, such as the Lost in the Middle (LiM) effect, where models perform better when information appears at the beginning (primacy bias) or end (recency bias) of the input, rather than in the middle. However, long-context studies have not consistently replicated these effects, raising questions about their intensity and the conditions under which they manifest. To address this, we conducted a comprehensive analysis using relative rather than absolute input lengths, defined with respect to each model's context window. Our findings reveal that the LiM effect is strongest when inputs occupy up to 50% of a model's context window. Beyond that, the primacy bias weakens, while recency bias remains relatively stable. This effectively eliminates the LiM effect; instead, we observe a distance-based bias, where model performance is better when relevant information is closer to the end of the input. Furthermore, our results suggest that successful retrieval is a prerequisite for reasoning in LLMs, and that the observed positional biases in reasoning are largely inherited from retrieval. These insights have implications for long-context tasks, the design of future LLM benchmarks, and evaluation methodologies for LLMs handling extended inputs.

As an essential part of modern hardware design, manually writing Register Transfer Level (RTL) code such as Verilog is often labor-intensive. Following the tremendous success of large language models (LLMs), researchers have begun to explore utilizing LLMs for generating RTL code. However, current studies primarily focus on generating simple single modules, which can not meet the demands in real world. In fact, due to challenges in managing long-context RTL code and complex cross-file dependencies, existing solutions cannot handle large-scale Verilog repositories in practical hardware development. As the first endeavor to exclusively adapt LLMs for large-scale RTL development, we propose RTLRepoCoder, a groundbreaking solution that incorporates specific fine-tuning and Retrieval-Augmented Generation (RAG) for repository-level Verilog code completion. Open-source Verilog repositories from the real world, along with an extended context size, are used for domain-specific fine-tuning. The optimized RAG system improves the information density of the input context by retrieving relevant code snippets. Tailored optimizations for RAG are carried out, including the embedding model, the cross-file context splitting strategy, and the chunk size. Our solution achieves state-of-the-art performance on public benchmark, significantly surpassing GPT-4 and advanced domain-specific LLMs on Edit Similarity and Exact Match rate. Comprehensive experiments demonstrate the remarkable effectiveness of our approach and offer insights for future work.

In this work, we present WLB-LLM, a workLoad-balanced 4D parallelism for large language model training. We first thoroughly analyze the workload imbalance issue in LLM training and identify two primary sources of imbalance at the pipeline parallelism and context parallelism levels. Then, to address the imbalance issue, at the pipeline parallelism level, WLB-LLM incorporates a workload-aware variable-length document packing method to balance the computation and communication workload across micro-batches. Additionally, at the context parallelism level, WLB-LLM introduces a novel fine-grained per-document sharding strategy, ensuring each worker within a context parallelism group has an identical workload. Comprehensive experiments under different model scales demonstrate that WLB-LLM significantly mitigates the workload imbalance during 4D parallelism LLM training and achieves an average speedup of 1.23x when applying WLB-LLM in our internal LLM training framework.

Recent advancements in LLM pretraining have featured ever-expanding context windows to process longer sequences. However, our pilot study reveals that models pretrained with shorter context windows consistently outperform their long-context counterparts under a fixed token budget. This finding motivates us to explore an optimal context window scheduling strategy to better balance long-context capability with pretraining efficiency. To this end, we propose SkyLadder, a simple yet effective approach that implements a short-to-long context window transition. SkyLadder preserves strong standard benchmark performance, while matching or exceeding baseline results on long-context tasks. Through extensive experiments, we pre-train 1B-parameter models (up to 32K context) and 3B-parameter models (8K context) on 100B tokens, demonstrating that SkyLadder yields consistent gains of up to 3.7% on common benchmarks, while achieving up to 22% faster training speeds compared to baselines. The evolution of language models has been marked by a consistent expansion in context window sizes (Figure 1 left). While early models like GPT (Radford, 2018) and BERT (Kenton & Toutanova, 2019) were limited to context windows of 512 tokens, subsequent models have pushed these boundaries significantly. GPT-2 (Radford et al., 2019) doubled this capacity to 1024 tokens, and with the advent of Large Language Models (LLMs) exceeding 1B parameters, the progression continued: Llama (Touvron et al., 2023a) implemented a 2048-token window, Llama-2 (Touvron et al., 2023b) extended it to 4096, and Llama-3 (Dubey et al., 2024) further expanded to 8192 tokens. The push to expand the context window is motivated by the need for models to handle longer sequences during inference. The development is also driven by a widespread belief that models pretrained with longer context windows should perform comparably to, or even surpass, their shorter context counterparts, as extended windows reduce document truncation and preserve coherence (Ding et al., 2024). We question whether the common belief that larger context windows does actually improve performance.

We present CoqPilot, a VS Code extension designed to help automate writing of Coq proofs. The plugin collects the parts of proofs marked with the admit tactic in a Coq file, i.e., proof holes, and combines LLMs along with non-machine-learning methods to generate proof candidates for the holes. Then, CoqPilot checks if each proof candidate solves the given subgoal and, if successful, replaces the hole with it. The focus of CoqPilot is twofold. Firstly, we want to allow users to seamlessly combine multiple Coq generation approaches and provide a zero-setup experience for our tool. Secondly, we want to deliver a platform for LLM-based experiments on Coq proof generation. We developed a benchmarking system for Coq generation methods, available in the plugin, and conducted an experiment using it, showcasing the framework's possibilities. Demo of CoqPilot is available at: https://youtu.be/oB1Lx-So9Lo. Code at: https://github.com/JetBrains-Research/coqpilot

Iterative refinement has emerged as an effective paradigm for enhancing the capabilities of large language models (LLMs) on complex tasks. However, existing approaches typically implement iterative refinement at the application or prompting level, relying on autoregressive (AR) modeling. The sequential token generation in AR models can lead to high inference latency. To overcome these challenges, we propose Context-Wise Order-Agnostic Language Modeling (COrAL), which incorporates iterative refinement directly into the LLM architecture while maintaining computational efficiency. Our approach models multiple token dependencies within manageable context windows, enabling the model to perform iterative refinement internally during the generation process. Leveraging the order-agnostic nature of COrAL, we introduce sliding blockwise order-agnostic decoding, which performs multi-token forward prediction and backward reconstruction within context windows. This allows the model to iteratively refine its outputs in parallel in the sliding block, effectively capturing diverse dependencies without the high inference cost of sequential generation. Empirical evaluations on reasoning tasks demonstrate that COrAL improves performance and inference speed, respectively, achieving absolute accuracy gains of $4.6\%$ on GSM8K and $4.0\%$ on LogiQA, along with inference speedups of up to $3.9\times$ over next-token baselines. Preliminary results on code generation indicate a drop in pass rates due to inconsistencies in order-agnostic outputs, highlighting the inherent quality--speed trade-off. Our code is publicly available at https://github.com/YuxiXie/COrAL.

We present MELODI, a novel memory architecture designed to efficiently process long documents using short context windows. The key principle behind MELODI is to represent short-term and long-term memory as a hierarchical compression scheme across both network layers and context windows. Specifically, the short-term memory is achieved through recurrent compression of context windows across multiple layers, ensuring smooth transitions between windows. In contrast, the long-term memory performs further compression within a single middle layer and aggregates information across context windows, effectively consolidating crucial information from the entire history. Compared to a strong baseline - the Memorizing Transformer employing dense attention over a large long-term memory (64K key-value pairs) - our method demonstrates superior performance on various long-context datasets while remarkably reducing the memory footprint by a factor of 8.

Supporting longer context for Large Language Models (LLM) is a promising direction to advance LLMs. As training a model for a longer context window is computationally expensive, many alternative solutions, such as Retrieval Augmented Generation (RAG), have been used. However, most existing RAG methods adopt embedding-based retrieval that falls short on long contexts. To address such challenges, we propose an attention-based retrieval technique, You Only Use Reactive Attention slice (YOURA). YOURA leverages a novel retrieval heuristic called reaction score to rank the relevance of each sentence in the input context with the query sentence. Intuitively, we measure how the per-token attention score "reacts" to the query and greedily retrieves the most reactive sentences. Internally, YOURA generates a token-indexed vector (called reaction vector) for the whole input context. To map each sentence to the token-indexed vector, we propose an Embedding-Agnostic Sentence Yield (EASY), a best-effort token wiggling algorithm. We evaluate our retrieval technique on three open-source pre-trained LLM models across six LongBench QA datasets. Our technique achieves up to 30% vLLM inference throughput improvement for serving long-context queries with a nearly identical quality score to the simple yet effective truncate-middle approach.