Subword tokenization impairs phonological knowledge encoding in LMs, but an IPA-based fine-tuning method restores it with minimal impact on other capabilities.
hub
Gonzalez, Hao Zhang, and Ion Stoica
11 Pith papers cite this work. Polarity classification is still indexing.
11
Pith papers citing it
hub tools
representative citing papers
SalesLLM provides an automatic evaluation framework for LLM sales dialogues that correlates 0.98 with human experts and shows top models approaching human performance while weaker ones lag.
Spoken language models exhibit style amnesia and fail to maintain instructed paralinguistic styles across multi-turn conversations, with explicit recall offering partial mitigation.
Evaluations of 53 LLMs on 14 basic math tasks show reasoning models use ~18x more tokens with sometimes lower accuracy, non-monotonic gains from extended budgets, and sharp performance drops under token constraints.
MARS² integrates multi-agent collaboration with tree-structured search in RL to boost code generation by increasing exploratory diversity and using path-level group advantages for credit assignment.
MOSAIC achieves mean macro F1 of 88 on chest X-ray report classification across five datasets in four languages using a 4B-parameter open model with low GPU memory and few-shot or light fine-tuning options.
LiveCodeBench collects 400 recent contest problems to create a contamination-free benchmark evaluating LLMs on code generation and related capabilities like self-repair and execution.
torchtune is a modular PyTorch library for LLM post-training that delivers competitive performance and memory efficiency while supporting rapid research iteration through hackable components.
SMCS coordinates 15 open-source LLMs via retrieval-based prior selection and exploration-exploitation posterior enhancement, outperforming GPT-4.1 by 5.36% and GPT-o3-mini by 5.28% on eight benchmarks.
Analysis of open-weight LLMs reveals strong English bias in generated sequences, with continual pre-training providing no cost benefit over from-scratch training for non-English adaptation.
SemanticQA unifies prior multiword expression datasets into a benchmark that reveals substantial performance variation among language models on semantic reasoning tasks.
citing papers explorer
-
How Tokenization Limits Phonological Knowledge Representation in Language Models and How to Improve Them
Subword tokenization impairs phonological knowledge encoding in LMs, but an IPA-based fine-tuning method restores it with minimal impact on other capabilities.
-
Sell More, Play Less: Benchmarking LLM Realistic Selling Skill
SalesLLM provides an automatic evaluation framework for LLM sales dialogues that correlates 0.98 with human experts and shows top models approaching human performance while weaker ones lag.
-
Style Amnesia: Investigating Speaking Style Degradation and Mitigation in Multi-Turn Spoken Language Models
Spoken language models exhibit style amnesia and fail to maintain instructed paralinguistic styles across multi-turn conversations, with explicit recall offering partial mitigation.
-
Do LLMs Overthink Basic Math Reasoning? Benchmarking the Accuracy-Efficiency Tradeoff in Language Models
Evaluations of 53 LLMs on 14 basic math tasks show reasoning models use ~18x more tokens with sometimes lower accuracy, non-monotonic gains from extended budgets, and sharp performance drops under token constraints.
-
MARS$^2$: Scaling Multi-Agent Tree Search via Reinforcement Learning for Code Generation
MARS² integrates multi-agent collaboration with tree-structured search in RL to boost code generation by increasing exploratory diversity and using path-level group advantages for credit assignment.
-
MOSAIC: A Multilingual, Taxonomy-Agnostic, and Computationally Efficient Approach for Radiological Report Classification
MOSAIC achieves mean macro F1 of 88 on chest X-ray report classification across five datasets in four languages using a 4B-parameter open model with low GPU memory and few-shot or light fine-tuning options.
-
LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code
LiveCodeBench collects 400 recent contest problems to create a contamination-free benchmark evaluating LLMs on code generation and related capabilities like self-repair and execution.
-
torchtune: PyTorch native post-training library
torchtune is a modular PyTorch library for LLM post-training that delivers competitive performance and memory efficiency while supporting rapid research iteration through hackable components.
-
A Scalable Multi-LLM Collaboration System with Retrieval-based Selection and Exploration-Exploitation-Driven Enhancement
SMCS coordinates 15 open-source LLMs via retrieval-based prior selection and exploration-exploitation posterior enhancement, outperforming GPT-4.1 by 5.36% and GPT-o3-mini by 5.28% on eight benchmarks.
-
Toward LLMs Beyond English-Centric Development
Analysis of open-weight LLMs reveals strong English bias in generated sequences, with continual pre-training providing no cost benefit over from-scratch training for non-English adaptation.
-
Revisiting a Pain in the Neck: A Semantic Reasoning Benchmark for Language Models
SemanticQA unifies prior multiword expression datasets into a benchmark that reveals substantial performance variation among language models on semantic reasoning tasks.