5arXiv cs.CL (Computation and Language)·
Self-Guided Test-Time Training improves long-context LLM utilization by up to 15%
Researchers propose Self-Guided Test-Time Training (S-TTT), a method that addresses the degradation in accuracy LLMs exhibit on long inputs by having the model first identify relevant evidence spans before applying test-time training only to those spans. A preliminary study on LongBench-v2 shows that TTT on randomly sampled spans hurts performance while TTT on oracle spans substantially helps, motivating the self-guided selection approach. Evaluated on LongBench-v2 and LongBench-Pro with Qwen3-4B-Thinking-2507 and Llama-3.1-8B-Instruct, S-TTT achieves up to 15% relative accuracy improvement. The method offers a practical path to better long-context utilization without the prohibitive cost of full-context adaptation.
Related guides (2)
Related events (8)
Test-Time Training End-to-End (TTT-E2E) Retrains Model Weights to Handle Long Inputs
Researchers from Astera Institute, Nvidia, Stanford, UC Berkeley, and UC San Diego introduced TTT-E2E, a method that compresses long context into transformer weights by training the model during inference via meta-learning. The approach uses sliding-window attention restricted to 8,000 tokens and updates only the fully connected layers of the last quarter of the network on each 1,000-token chunk at inference time, keeping per-token generation latency roughly constant as context scales to 128,000 tokens. TTT-E2E slightly outperforms vanilla transformers on next-token prediction loss across long contexts and matches efficient architectures like Mamba 2 and Gated DeltaNet on inference speed, but fails dramatically on Needle-in-a-Haystack retrieval beyond 8,000 tokens and incurs substantially higher training latency. The work reframes long-context handling as a training-inference trade-off rather than an architectural design problem.
ReContext: Training-free recursive evidence replay improves LLM long-context reasoning
Researchers introduce RECONTEXT, a training-free inference-time method for improving long-context reasoning in LLMs. The approach uses model-internal relevance signals to build a query-conditioned evidence pool that is replayed before final generation, without modifying the original context, external memory, or context pruning. Experiments across eight long-context datasets at 128K context length show consistent improvements on Qwen3-4B, Qwen3-8B, and Llama3-8B. The authors provide a theoretical grounding via associative memory theory, framing attention as cue-trace association and replay as trace reactivation.
NLL-guided training-free method selects optimal full-attention layers for efficient long-context inference
Researchers propose NLL-guided layer selection, a training-free technique for hybrid attention models that identifies which layers should use full versus sliding-window attention by measuring negative log-likelihood degradation on answer tokens. On LongMemEval with Qwen3-4B, the method achieves 64.6% accuracy using only 1/4 full-attention layers, matching a 1/2-FA periodic baseline while halving compute, and outperforming a periodic 1/4-FA baseline by 10.4 percentage points. The calibration procedure requires approximately 15 minutes of one-time compute, making it practical for deployment. The work advances the efficiency-accuracy tradeoff for long-context LLM inference without requiring any retraining.
LongCrafter: Evidence-graph-guided synthesis framework for long-context SFT data
LongCrafter is a structured framework for synthesizing long-context supervised fine-tuning data, addressing limitations of prior approaches including narrow task coverage, low difficulty, and lack of faithfulness supervision. The system uses a hierarchical 32-task taxonomy and constructs explicit evidence graphs modeling cross-paragraph dependencies to generate grounded instruction-response pairs. Models fine-tuned on LongCrafter data outperform SFT baselines and official post-trained models on LongBench, LongBench v2, and LooGLE for both Qwen2.5-7B and LLaMA-3.1-8B, with notable gains on high-difficulty tasks and improved robustness to the 'lost in the middle' problem.
Test-time scaling for small VLMs on multilingual visual MCQ: conditions matter more than methods
A new arXiv paper examines whether test-time scaling (TTS) transfers to small open vision-language models using EXAMS-V, a multilingual visual multiple-choice benchmark. The study compares self-consistency, describe-then-reason with PRM-guided beam search, and post-hoc selectors across Qwen2.5-VL-7B-Instruct and Qwen3.5-4B. Key findings: prompt parseability and decoding budget (token limit) dominate gains, while elaborate search/verification methods like PRM-guided beam search underperform plain majority vote at 8x the cost. The best configuration achieves 84.1% on ImageCLEF 2026 test split, ranking first on the Visual MCQ leaderboard.
HullFT: Efficient Test-Time Finetuning of LLMs via Convex Reconstruction and Gradient Caching
HullFT is a new method for test-time finetuning (TTFT) of language models that addresses the dual bottlenecks of retrieval quality and per-query finetuning cost. It represents query embeddings as sparse convex combinations of training sequences using Frank-Wolfe optimization, yielding diverse and relevant support sets without expensive diversity-aware search. A geometric integerization step converts fractional weights into integer multiplicities, enabling a Gradient Reuse scheme that amortizes forward-backward computation across repeated examples. Experiments show improved quality-efficiency tradeoffs over prior TTFT methods, measured in bits-per-byte at lower total runtime.
Towards Reliable Multilingual LLMs-as-a-Judge: An Empirical Study
This paper systematically investigates strategies for extending LLM-based automatic evaluation (LLMs-as-a-Judge) to multilingual settings, covering high-, mid-, and low-resource languages (English, Spanish, Basque). The authors compare instruction translation, monolingual vs. multilingual supervision, and model size, finding that fine-tuned smaller models can match proprietary models when in-domain data is available, while zero-shot larger models are preferable out-of-domain. Two meta-evaluation datasets are extended to Spanish and Basque, and all data and code are publicly released.
CLP: Lightweight collocation-length predictor achieves zero-loss multi-token inference speedup
Researchers propose CLP (Collocation-Length Predictor), a span-level decision layer for accelerating LLM inference via multi-token prediction without quality degradation. The key insight is 'Backbone-as-Architect': the backbone LM head always generates the first token while MTP heads handle only subsequent tokens, eliminating head-backbone competition that causes repetitive outputs in prior methods. CLP uses a single linear layer (~4.6K–7.7K parameters) versus 1M-parameter gate networks in prior work, achieving 1.14x–1.29x speedup on Qwen2.5 models with near-zero repetition ratio. The paper also establishes that shorter prediction horizons improve MTP head accuracy on larger models, offering a scaling-aware design principle.