ICML 2026 Papers

As real-world tasks grow increasingly complex, long-context reasoning has become a core capability for Large Language Models (LLMs). However, few studies explore which data types are effective for long-context reasoning and why. We find that structured table data with periodic structures shows strong potential for long-context reasoning. Motivated by this observation, we mathematically analyze tabular dependency structures using mutual information, revealing periodic non-vanishing dependencies in table data. Furthermore, we systematically analyze the capabilities of structured table data, conduct relevant scaling experiments, and validate its underlying mechanisms for enhancing long-context reasoning, yielding several meaningful insights. Leveraging these insights, we propose a simple yet scalable pipeline(TableLong) for synthesizing high-quality, diverse, and verifiable structured table data to boost long-context reasoning via RL. Extensive experimental results demonstrate that table data significantly enhances the long-context reasoning capability of LLMs across multiple long-context benchmarks (+8.24\% on average), and even improves performance on out-of-domain benchmarks (+8.06\% on average). We hope that our insights provide practical guidance for effective post-training data to enhance long-context reasoning in LLMs.

Large language model (LLM) exists a subset of attention heads that are highly responsible for long-context processing. Existing work has identified different long-context heads in models, but their detection methods mainly rely on model inference on actual long texts and do not analyze the inherent properties of the head parameters. In this paper, we use kernel methods to analyze static \textit{frequency kernels} formed by different rotation frequency components of attention heads, and we design a Long-context Potential Score (LPS) to measure the potential of attention heads in processing long contexts. Kernels of heads with high LPS exhibit concentrated low-frequency energy and low effective rank, which allow them to effectively capture highly specialized information from distant contexts. Experiments and analysis on long-context tasks and model behaviors show that the LPS metrics can well reflect the actual capability of heads on long contexts. Furthermore, by simply amplifying low-frequency kernels of heads with high retrieval potential, we can further improve model's performance on long-context tasks. Our metrics and head enhancement methods are fully static and offline, and they can be quickly conducted under low-resource constraints. Code is publicly available at [here](https://github.com/OpenDFM/long-context-heads).

Tool calling extends large language models (LLMs) by enabling grounded interaction with external executable interfaces, thereby supporting environment-coupled problem solving. However, mainstream in-context learning (ICL) approaches typically incorporate detailed tool documentation and usage examples directly into the context. This results in substantial inference overhead and heightened risks of hallucination as the context length grows. Conversely, while tuning-based methods improve general tool-calling capabilities, they often fail to effectively internalize the specific details of previously seen tools, thereby retaining a dependency on in-context documentation. To address these limitations, we propose ParaTool, a framework that projects each tool into a dedicated, loadable set of parameters. By equipping a dynamic integration of these parameterized tools, the LLM can perform tool calling without relying on in-context documents or examples. Specifically, our approach consists of three stages: (1) parametric tool pre-training encapsulates the knowledge of different tools into independent parameter modules; (2) soft tool selection employs a gating network to dynamically weigh and aggregate relevant tool parameters; and (3) parametric tool fine-tuning jointly updates tool parameters to align the training and inference processes. Experiments on Stable ToolBench and BFCL demonstrate that ParaTool significantly outperforms strong ICL-based baselines, achieving superior performance while reducing computational complexity.

Standard softmax attention scales quadratically with sequence length, which makes long-context training and inference expensive. We introduce PLASH, an attention block whose cost grows linearly in the number of keys. PLASH compresses the original keys and values into $M$ learned prototypes, where $M\in\mathbb{Z}_{>0}$ is much smaller than the number of keys. The compressed prototypes are then enriched with randomized polynomial features that recover inter-token information lost to compression. The output is computed by exact scaled dot-product softmax attention from each query to the enriched prototypes, so PLASH preserves the standard attention interface.The construction applies to self- and cross-attention. We prove sketch-error bounds for the enrichment step, a per-input certificate that upper-bounds the deviation from standard softmax attention on each forward pass, and a runtime bound linear in the number of queries and keys. Experiments on long-context language modeling (Qwen3-4B on PG-19) and time-series forecasting (ETT, ECL, Weather) show competitive accuracy and favorable scaling against efficient-attention baselines.

The quadratic complexity of attention remains the central bottleneck in long-context inference for large language models. Prior acceleration methods either sparsify the attention map with structured patterns or permanently evict tokens at specific layers, which can retain irrelevant tokens or rely on irreversible early decisions despite the layer-/head-wise dynamics of token importance. In this paper, we propose Token Sparse Attention, a lightweight and dynamic token-level sparsification mechanism that compresses per-head $Q, K, V$ to a reduced token set during attention and then decompresses the output back to the original sequence, enabling token information to be reconsidered in subsequent layers. Furthermore, Token Sparse Attention exposes a new design point at the intersection of token selection and sparse attention. Our approach is fully compatible with dense attention implementations, including Flash Attention, and can be seamlessly composed with existing sparse attention kernels. Experimental results show that Token Sparse Attention consistently improves accuracy–latency trade-off, achieving up to $\times$3.23 attention speedup at 128K context with less than 1\% accuracy degradation. These results demonstrate that dynamic and interleaved token-level sparsification is a complementary and effective strategy for scalable long-context inference.

Induction heads are attention circuits believed to underlie in-context learning in transformers, yet a precise characterization of the estimators they implement remains elusive. We study transformers trained on order-$k$ Markov chains and identify two complementary smoothing mechanisms. First, at finite attention-weight scale, the circuit implements a soft context-matching estimator: it aggregates contributions from exact and partial context matches, weighted exponentially by their overlap, and induces a data-dependent interpolation across context orders analogous to Jelinek--Mercer smoothing. Second, a beginning-of-sequence (BOS) token induces additive pseudo-counts, recovering Dirichlet-style smoothing. We construct a disentangled transformer implementing both mechanisms and show that trained transformers recover the predicted attention patterns. Across settings where pseudo-count smoothing is optimal or lower-order contexts provide structured evidence, trained transformers match or outperform classical count-based baselines. Our results bridge mechanistic interpretability of induction heads with classical statistical smoothing, revealing that transformers learn to regularize in-context estimation rather than simply count.

We propose ContextLM, a framework that implicitly learns multi-token prediction by augmenting standard pretraining with an intrinsic next-context prediction objective. ContextLM builds a language model on top of context embeddings that span multiple tokens, enabling better next-token prediction by predicting the next context. Our model is fully compatible with standard autoregressive, token-by-token evaluation paradigms (e.g., perplexity). Extensive experiments with GPT-2 and Pythia backbones (up to 1.5B parameters and 300B training tokens) reveal that ContextLM shifts the Pareto frontier of scaling laws, exhibiting superior efficiency in parameters, training tokens, and FLOPs. Our results show that ContextLM could already achieve the baseline perplexity using 39\% fewer parameters and demonstrates robust generalization improvements on extensive downstream tasks under equivalent parameter counts.

Large language models (LLMs) increasingly operate in settings that require reliable long-context understanding, such as retrieval-augmented generation and multi-document reasoning. A common strategy is to fine-tune pretrained short-context models at the target sequence length. However, we find that standard long-context adaptation can remain brittle: model accuracy depends strongly on the absolute placement of relevant evidence, exhibiting high positional variance even when controlling for task format and difficulty. We propose *RoPE-Perturbed Self-Distillation*, a training regularizer that improves positional robustness. The core idea is to form alternative ``views'' of the same training sequence by perturbing its RoPE indices---effectively moving parts of the context to different positions---and to train the model to produce consistent predictions across views via self-distillation. This encourages reliance on semantic signals instead of brittle position dependencies. Experiments on long-context adaptation of Llama-3-8B and Qwen-3-4B demonstrate consistent gains on long-context benchmarks, including up to 12.04 percent-point improvement on RULER-64K for Llama-3-8B and 2.71 percent-point gain on RULER-256K for Qwen-3-4B after SFT, alongside improved length extrapolation beyond the training context window.

Effective demonstration selection is crucial for maximizing large language model (LLM) performance in few-shot in-context learning. Because of effects such as recency bias, the effectiveness of demonstrations depends heavily on their contextual relationship to the specific query and on the ordering in which they are presented, making demonstration selection a complex combinatorial problem. To address these two challenges, we introduce AutoSelect, a novel framework that formulates demonstration selection as an auto-regressive sequential decision process. At each step, AutoSelect embeds the query and previously selected demonstrations into matrix representations to preserve structural information, and a trainable policy model sequentially selects the next best exemplar. To navigate the factorial space of demonstration permutations, our framework formulates a Kullback-Leibler (KL)-regularized optimization problem, from which an optimal policy induces an optimal Plackett-Luce (PL) ranking over all possible demonstration sequences. We prove that minimizing a tractable policy-level cross-entropy (CE) loss provably bounds the worst-case discrepancy between our policy's induced PL ranking and the optimal one, enabling tractable prioritization of high-quality sequences. Empirically, AutoSelect outperforms existing heuristic and learning-based methods across nine diverse datasets, achieving up to an 11\% improvement over the strongest baseline. Analytical studies and a case study further highlight AutoSelect's key properties, as well as its transferability and generalizability.

Sparse attention has been proposed as a way to alleviate the quadratic cost of transformers, a central bottleneck in long-context training. A promising line of work is $\alpha$-entmax attention, a differentiable sparse alternative to softmax that enables input-dependent sparsity yet has lagged behind softmax due to the computational overhead necessary to compute the normalizer $\tau$. In this paper, we introduce AdaSplash-2, which addresses this limitation through a novel histogram-based initialization that reduces the number of iterations needed to compute $\tau$ to typically 1-2. The key idea is to compute a coarse histogram of attention scores on the fly and store it in on-chip SRAM, yielding a more accurate initialization that enables fast forward and backward computation. Combined with a sparsity-aware GPU implementation that skips zero blocks with low overhead, AdaSplash-2 matches or improves per-step training time relative to FlashAttention-2 when block sparsity is moderate-to-high (e.g., $>$60\%), which often occurs at long-context lengths. On downstream tasks, models trained with our efficient $\alpha$-entmax attention match softmax baselines at short-context lengths and achieve substantial gains in long-context settings.

Efficiently processing long sequences with Transformer models usually requires splitting the computations across accelerators via context parallelism. The dominant approaches in this family of methods, such as Ring Attention or DeepSpeed Ulysses, enable scaling over the context dimension but do not focus on memory efficiency, which limits the sequence lengths they can support. More advanced techniques, such as Fully Pipelined Distributed Transformer or activation offloading, can further extend the possible context length at the cost of training throughput. In this paper, we present UPipe, a simple yet effective context parallelism technique that performs fine-grained chunking at the attention head level. This technique significantly reduces the activation memory usage of self-attention, breaking the activation memory barrier and unlocking much longer context lengths. Our approach lowers the peak activation memory usage by as much as **82.5%** for 70B Transformers, while matching previous context parallelism techniques in terms of training speed. UPipe can support maximum context lengths of up to 5M tokens for training 8B models on a single 8xH100 node, improving upon prior methods by **25%**.

In-context learning (ICL) derives its power from enabling Large Language Models to adapt to new tasks via prompt-based reasoning alone, entirely bypassing the need for parameter updates. Existing theories primarily study ICL in single-task settings, while real-world prompts often contain sequences of heterogeneous tasks, leaving a gap in understanding whether Large Language Models implicitly perform continual learning during inference. To bridge this gap, we propose the first theoretical framework for in-context continual learning, modeling how a pretrained Transformer processes multiple sequential tasks within a single prompt through shared attention mechanisms. Focusing on linear and masked linear self-attention, we derive error expressions for model predictions under sequential task prompts and analyze their generalization and forgetting behavior. Our results reveal that standard attention mechanisms inevitably induce inter-task interference by uniformly or causally aggregating historical contexts, leading to systematic bias. We further provide a bias–variance–interference decomposition of prediction error, characterizing when historical in-context information yields positive transfer or provable negative transfer. This analysis exposes fundamental limits of attention-based continual inference and offers theoretical explanations for order sensitivity and performance degradation in long prompts.

Large language models (LLMs) are increasingly deployed as agents in dynamic real-world environments, where success depends on maintaining precise records of actions and observations. However, the resulting unbounded context growth in long-horizon agentic tasks makes two critical bottlenecks: prohibitive inference memory costs and reasoning degradation due to irrelevant information. Existing compression methods fail to fully address this, often relying on brittle heuristics or requiring parameter updates impractical for proprietary or large-scale LLMs. We introduce Agent Context Optimization (ACON), a unified framework that optimally compresses both observations and history into concise, informative representations. Distinct from prior works, ACON employs an optimization in natural language space: it iteratively refines compression guidelines based on failure analysis of the agent, ensuring critical state information is preserved without model fine-tuning. To further minimize computational overhead, we distill the optimized compressor into smaller models. Experiments on AppWorld, OfficeBench, and Multi-objective QA demonstrate that ACON reduces peak token usage by 26–54\% while improving task success over existing compression baselines. Notably, it enables smaller LMs to function effectively as long-horizon agents, achieving up to 46% performance improvement by mitigating context distraction. Our code is available at https://github.com/microsoft/acon.

Generating long-form content, such as minute-long videos and extended texts, is increasingly important for modern generative models. Block diffusion improves inference efficiency via KV caching and block-wise causal inference and has been widely adopted in diffusion language models and video generation. However, in long-context settings, block diffusion still incurs substantial overhead from repeatedly computing attention over an ever-growing KV cache. We identify an underexplored property of block diffusion: cross-step redundancy of attention within a block. Our analysis shows that attention outputs from tokens outside the current block remain largely stable across diffusion steps, while block-internal attention varies significantly. Based on this observation, we propose FlashBlock, a cached block-external attention mechanism that reuses stable attention output, substantially reducing attention computation and KV cache access without modifying the diffusion process. Moreover, FlashBlock is orthogonal to sparse attention and can be combined as a complementary residual reuse strategy. When integrated, it substantially improves model accuracy under aggressive sparsification by offsetting much of the performance loss induced by sparsity. Experiments on diffusion language models and video generation demonstrate up to 1.44$\times$ higher token throughput and up to 1.6$\times$ reduction in attention time, with negligible impact on generation quality.

Evaluation of long-context Large Language Models (LLMs) has advanced rapidly. However, most existing benchmarks are limited to the document level and focus mainly on high-resource languages, leaving many fine-grained challenges insufficiently evaluated. To address this gap, we present MGAL, the first multilingual, granularity- and position-aware long-context benchmark. MGAL is constructed from United Nations (UN) reports spanning 8K to 128K tokens across the six official UN languages. It covers four coherent levels of linguistic granularity (word, sentence, paragraph, and document) and further stratifies entries by their position within the document (begin, middle, and end), indexed at both the document and paragraph levels. This design enables systematic diagnosis of multilingual long-context comprehension across different granularities. Through extensive experiments and analyses, we find that: (1) LLMs perform well at word-level tasks but struggle with coarser-grained ones; and (2) Closed-source models retain a clear performance advantage in lower-resource languages. We further identify two new challenges: (1) Under local semantic crowding, where neighboring sentences share topics and entities, models tend to follow surface cues (e.g., connectives like 'however' or repeated entities) rather than the discourse role of the sentence in surrounding context (e.g., background, outcome); and (2) A gap between fluency and consistency in generated outputs, where models produce text that reads smoothly but drifts from the source facts. In addition, we observe several patterns in line with prior studies, including reliance on nearby evidence and reuse of options under uncertainty.

Human episodic memory supports the retrieval of experiences that unfold over extended timescales, yet the computational mechanisms underlying this ability remain debated due to the limited mechanistic accessibility in long-term memory experiments in humans. Long-context LLMs may offer promising ways to reveal plausible computational mechanisms that drive this type of retrieval. Here, we investigate whether and how LLMs capture the core behavioral signatures of episodic memory via a temporal order memory task. Using a new dataset of human behavior based on memory of a full-length novel, we show that models exhibit the same characteristic distance effect observed in humans on this task. We next apply long-context mechanistic interpretability analyses to uncover how models solve this task, and find that model performance relies on a one-dimensional temporal code that is reinstated during retrieval by a single time-reinstatement attention head. These findings support temporal context reinstatement as an important mechanism for episodic-like temporal-order memory in LLMs, offering new insights into how temporal aspects of long-term episodic memory may be instantiated in both artificial and biological systems.

Large Language Models (LLMs) are fundamentally limited by representation collapse, a bottleneck that severely degrades long-context performance. We identify that existing approaches risk drifting into one of two pathological extremes: homogenization collapse (e.g., attention sinks causing rank deficiency) and isolation collapse (e.g., local attention causing context disconnection). Through spectral analysis of attention dynamics, we derive an intrinsic trade-off between mixing efficiency (spectral gap) and information capacity (effective rank) that standard mechanisms struggle to balance. To resolve this dilemma, we propose the Topologically Regularized Side-Path (TRSP), a non-invasive architectural intervention that achieves spectral balance. TRSP employs a parameter-free Triangular Box mechanism, scaled by a lightweight, length-aware gate, to regularize the token interaction topology. By integrating proximal coupling to preserve effective rank and distal propagation to support non-degenerate mixing, TRSP promotes a geometrically healthier transition operator without altering core attention. Experiments show significant improvements across general capabilities and long-context benchmarks. Notably, on NoLiMa at $8\times$ the training length, TRSP retains $83\%$ accuracy and surpasses the Differential Transformer and Gated Attention by approximately 30 and 50 percentage points, respectively. Code available at: https://github.com/Eziotao-tyd/TRSP.