arXiv AI
Jun 18
Rescaling MLM-Head for Neural Sparse Retrieval
arXiv:2606. 18811v1 Announce Type: cross Abstract: Learned sparse retrieval (LSR) models such as SPLADE have traditionally used BERT-style masked language models as backbone encoders.
Why Advanced Encoders Lag on Sparse Retrieval? The Answer and an Approach to Bridging Vocabulary Gaps
arXiv:2607. 00004v1 Announce Type: cross Abstract: While advanced foundation models like ModernBERT significantly outperform older architectures in dense retrieval, they surprisingly lag behind the aging BERT-base baseline in learned sparse retrieval (LSR).
Related stories
arXiv AI
Jul 14
Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models
arXiv:2601. 07372v2 Announce Type: replace-cross Abstract: While Mixture-of-Experts (MoE) scales capacity via conditional computation, Transformers lack a native primitive for knowledge lookup, forcing them to inefficiently simulate retrieval through computation.
arXiv AI
Jun 4
L$^3$: Large Lookup Layers
arXiv:2601. 21461v3 Announce Type: replace-cross Abstract: Modern sparse language models typically achieve sparsity through Mixture-of-Experts (MoE) layers, which dynamically route tokens to dense MLP "experts.
arXiv AI
Aug 11
LF${}^{2}$AR: Accounting for Layerwise Dynamics to Improve Multimodal Adaptation of Language Models
arXiv:2503. 06211v3 Announce Type: replace-cross Abstract: Text-pretrained language models (LMs) encode rich world knowledge, but adapting them to process and generate perceptual modalities such as audio and images while effectively leveraging that knowledge remains challenging.
arXiv AI
Jul 17
In-Place Tokenizer Expansion for Pre-trained LLMs
arXiv:2607. 15232v1 Announce Type: cross Abstract: A tokenizer fixed at the start of pre-training allocates vocabulary in proportion to the pre-training corpus, reflecting the deployment priorities at that time.
arXiv Machine Learning
Jun 2
Test-Time Compute for Frozen Embedding Models through Agentic Program Search
arXiv:2605. 11374v5 Announce Type: replace Abstract: Test-time compute is widely believed to benefit only large reasoning models, leaving small models with nothing to gain.
arXiv Machine Learning
Jul 24
Adaptive Depth Sparse Framework: Similarity-Driven Resource Allocation for Pre-Trained LLMs
arXiv:2607. 21291v1 Announce Type: cross Abstract: Large language models (LLMs) achieve strong generation and reasoning performance, but the Transformer architecture incurs high inference cost.
arXiv AI
Jun 6
Predict and Reconstruct: Joint Objectives for Self-Supervised Language Representation Learning
arXiv:2606. 05173v1 Announce Type: cross Abstract: Masked language modelling (MLM) has been the dominant pre-training objective for text encoders since BERT, yet it encourages representations that are strongly anchored to surface-form token identity rather than deeper semantic structure.
arXiv AI
Jun 9
Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps
arXiv:2605. 16928v2 Announce Type: replace-cross Abstract: Long-context inference in large language models is bottlenecked by the quadratic cost of full attention.
arXiv Machine Learning
Aug 5
Beyond Initialization Loss: A Systematic Study of Token Embedding Initialization Strategies for LLM Vocabulary Extension
arXiv:2608. 03494v1 Announce Type: cross Abstract: Vocabulary extension is an efficient way to adapt pretrained large language models (LLMs) to new languages, but the initialization of newly added token embeddings can strongly affect continued pre-training (CPT) efficiency.
arXiv AI
Jun 29
VASAE: Naming SAE Dictionary Directions with Vocabulary-Aligned Anchoring
arXiv:2606. 27941v1 Announce Type: cross Abstract: Sparse autoencoders (SAEs) provide useful decompositions of Transformer residual streams, but their learned features are usually named post hoc rather than directly connected to the Transformer's token vocabulary.
arXiv Machine Learning
Aug 6
Kathleen Writes: Autoregressive Generation and Data Scaling Without Attention
arXiv:2608. 04678v1 Announce Type: cross Abstract: Papers 1-2 of the Kathleen series showed that a byte-level, attention-free architecture built from a wavetable encoder and multi-scale reverberant state can match strong baselines on classification at ~450-700K parameters, without pretraining.