HAKARI-Bench reconstructs 35 benchmarks into 551 tasks across 43 languages, reproducing full MTEB, MMTEB, and BEIR rankings with Spearman correlation above 0.97 while supporting efficiency variant comparisons.
Under review
9 Pith papers cite this work. Polarity classification is still indexing.
9
Pith papers citing it
years
2026 9representative citing papers
A temporary CLM phase followed by MLM decay during encoder continued pretraining outperforms standard MLM on biomedical tasks by 0.3-2.8pp across languages and model sizes.
A probabilistic model with domain-aligned inductive bias detects acts of mechanistic reasoning in student conversations and shows improved generalization to unseen students and novel contexts.
MADE creates a contamination-resistant living benchmark for multi-label classification of medical device adverse events, with evaluations revealing model-specific trade-offs in accuracy and uncertainty quantification.
Cross-encoder reranker performance scales predictably via power laws with model size and training exposure, allowing accurate forecasts for 400M and 1B models and data-heavy compute allocation.
Fine-tuned ModernBERT-family encoders match LLM judges on F1, false negative rate, and precision-recall for harmful output detection across adversarial datasets and attack types while promising lower cost and latency.
Rescaling the MLM-head projection by a constant factor at initialization resolves scale mismatch in learned sparse retrieval and enables stable use of large-norm backbones such as ModernBERT.
Modestly sized language models acquire sensitivity to the meanings of rare Paired-Focus constructions later than their syntactic forms, with semantic learning correlating to gains in selected world-knowledge domains.
MATCH augments sparsified attention with an efficient in-context retrieval system to boost performance on long-range recall tasks in transformers.
citing papers explorer
-
A Causal Language Modeling Detour Improves Encoder Continued Pretraining
A temporary CLM phase followed by MLM decay during encoder continued pretraining outperforms standard MLM on biomedical tasks by 0.3-2.8pp across languages and model sizes.