4arXiv cs.CL (Computation and Language)·
Controlled factorial study disentangles architecture, model variant, and scale effects in LLM-based entity matching
A new arXiv preprint presents a controlled factorial study of language model-based entity matching across three matcher architectures (bi-encoder, cross-encoder, generative), three model variants, and three model sizes from the Qwen3 family, totaling 1,215 fine-tuning runs on nine datasets. Key findings include: model variant (pretraining objective) is more important than scale for bi-encoders; cross-encoders consistently outperform bi-encoders but larger models narrow the gap; generative matchers only outperform cross-encoders under distribution shift; and larger models are more prone to shortcut learning. The study also evaluates cross-dataset transferability and computational cost, releasing all code and results.
Related guides (3)
Related events (8)
Sub-billion parameter SLMs outperform zero-shot GPT-5.4 and Claude Sonnet 4.6 on relation extraction benchmarks
A new arXiv paper demonstrates that small language models (360M–3B parameters) fine-tuned on task-specific data can substantially outperform zero-shot frontier LLMs on relation extraction tasks. The best sub-billion model, Qwen2.5-0.5B fine-tuned on pooled general-domain data, achieves micro-F1 of 0.83 versus 0.69 for GPT-5.4 and 0.66 for Claude Sonnet 4.6 in zero-shot settings. The authors attribute the gains to task adaptation rather than model architecture, with a discriminative RoBERTa baseline also exceeding frontier models, and show that 4-bit quantized models deployable on consumer GPUs can match or beat proprietary API-based systems for this narrow task. The work provides evidence that for well-defined NLP tasks with available training data, compact adapted models offer a practical, private, and hardware-efficient alternative to frontier APIs.
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.
Multilingual study finds LLMs are not uniformly robust to non-canonical tokenizations, with up to 23.7% performance drops
A new arXiv paper investigates how language models behave when given alternative (non-canonical) tokenizations of the same input string across 27 languages and six downstream tasks. While prior work showed English models are largely invariant to such perturbations, the study finds this does not generalize: Llama-3.1-8B drops 23.7% on average, Qwen3-8B 11.4%, and Gemma-3-12B 9.9% in relative performance. Languages with higher token fragmentation are systematically more sensitive, and the authors show LoRA fine-tuning on multi-tokenization data—including English-only data—provides meaningful mitigation.
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.
Scaling laws for native multimodal pre-training from scratch characterized empirically
A new arXiv preprint investigates compute-optimal scaling for transformer-based vision-language models trained natively on multimodal inputs from scratch, rather than via late-fusion of separately pre-trained components. The authors derive power-law relationships between compute budget, model size, and token count, finding that language and multimodal objectives exhibit distinct scaling behaviors and that data mixture composition strongly influences optimal resource allocation. They also report positive cross-modal transfer effects, including improved pure-text spatial reasoning. The work establishes an efficiency frontier for configuring model size, token count, and data mixture under fixed compute.
Instruction-tuned LLMs show stronger syntactic convergence toward human interlocutors than humans do
A new arXiv preprint measures syntactic convergence in 16 open-weight Llama and Gemma models (1B–70B, pretrained and instruction-tuned) by substituting model outputs into pre-existing human dialogues and measuring context-free grammar rule reuse. Every model showed greater CFG-rule overlap with the preceding human turn than a random baseline, and instruction-tuned models exceeded the syntactic convergence of the human responses they replaced. Instruction tuning increased local syntactic and semantic alignment with interlocutors, though it also raised baseline overlap with unrelated primes, suggesting a more globally accommodating output style rather than purely targeted priming.
Graded entity-familiarity probes in LLMs enable refusal steering and cross-language robustness
Researchers probe activations at the final prompt token across twelve instruction-tuned models (Bielik, PLLuM, Gemma-4, Qwen3 families) to assess whether LLMs encode graded familiarity with named entities before generating answers. Using a new dataset of 1,440 Polish entities across popularity deciles plus fabricated controls, they find familiarity probes reliably separate real from fabricated entities and track popularity in Polish-adapted models. A key finding is that a single one-dimensional familiarity direction injected at one layer in Gemma-4-12B can steer refusal rates monotonically from 0.24 to 1.00 (or 0.73 to 0.00), revealing a separation between representational familiarity and the policy that converts it into abstention behavior.
Investing in Performance: Fine-tune small models with LLM insights — a CFM case study
This Hugging Face blog post presents a case study from CFM (Capital Fund Management) on using large language model outputs to guide fine-tuning of smaller, more efficient models for financial applications. The approach leverages LLM-generated signals or labels to train compact models that can be deployed at lower cost and latency. The case study illustrates an enterprise pattern of distilling LLM capabilities into task-specific smaller models for production use.