Notable data and AI papers

Papers627Labs45

627 papers · 45 labs

When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation research paper by Microsoft, 2026

Microsoft · Training and scaling · Sep 17, 2026 · 6 citations · 109 upvotes · unverified

Read on arXiv

What it shows

This paper identifies termination-token mismatch as a cause of length inflation in policy distillation and provides a mitigation strategy.

By Yuxiao Yang, Tianrun Yu, Shangzhe Li and 6 more · arXiv 2609.20511 · PDF · Code

UnverifiedSummarised by a model from the abstract; not yet checked by hand.

More from Microsoft

All 57
Agents and evaluationUnverifiedNov 2, 2026WebHarbor: Docking Real Websites for Evolving GUI Agent EnvironmentsWebHarbor provides realistic, containerized website environments for training and testing computer-using web agents across complex online tasks.Microsoft28 upvotesRead →Agents and evaluationUnverifiedOct 19, 2026Agent World Model in the Wild: Computer-Use Agent Environment SynthesisThis pipeline creates controllable, open-source environments to safely train and test AI agents on complex, long-horizon computer-use tasks.Microsoft69 upvotesRead →Training and scalingUnverifiedOct 16, 2026Self-Evolving Coding Agents via Harness-Policy OptimizationThis framework improves coding agents by recursively alternating between optimizing the agent's policy and the environment harness it uses to interact with code.Microsoft48 upvotesRead →Multimodal and roboticsUnverifiedOct 13, 2026AesCode: Aesthetic Code Generation with Decoupled Cross-Modal RewardsThis paper introduces a method to improve code-based visual generation by using image-based aesthetic guidance and decoupled rewards to ensure quality and correctness.Microsoft137 upvotesRead →Training and scalingUnverifiedOct 13, 2026APT: Feedback-Driven Agentic Post-Training for Text-to-Image GenerationThis paper introduces an automated post-training framework that uses a feedback loop of evaluation and diagnosis to improve text-to-image models.Microsoft23 upvotesRead →Multimodal and roboticsUnverifiedOct 13, 2026LensErase: A Foundation Model for Erase AnythingThis paper introduces a foundation model designed for versatile image editing and object removal using mask-based inputs and efficient positional encoding.Microsoft52 upvotesRead →Inference and efficiencyUnverifiedOct 13, 2026Simplifying and Improving On-Policy Distillation for Few-Step DiffusionThis paper introduces a simplified regression-based method for distilling diffusion models that improves generation speed without requiring complex training techniques.Microsoft27 upvotesRead →Applied AIUnverifiedOct 5, 2026MiniCorp: The Last Mile of the AI Agent FirmMiniCorp is an office simulator that allows researchers to study how agents run companies and generate enterprise data for training.Microsoft39 upvotesRead →

Other training and scaling papers

Topic
Training and scalingJan 23, 2020Scaling Laws for Neural Language ModelsLanguage model loss falls as a smooth power law as model size, data and compute grow.OpenAI9,592 citations12 upvotesRead →Training and scalingMar 29, 2022Training Compute-Optimal Large Language ModelsChinchilla: for a fixed compute budget, train a smaller model on more data; parameters and tokens should grow together.Google DeepMind4,035 citations13 upvotesRead →Training and scalingSep 17, 2019Megatron-LM: Training Multi-Billion Parameter Language Models Using Model ParallelismMegatron-LM splits each Transformer layer across GPUs to train models with billions of parameters.NVIDIA3,387 citations5 upvotesRead →Training and scalingUnverifiedJan 5, 2024DeepSeek LLM: Scaling Open-Source Language Models with LongtermismDeepSeek LLM, an open-source language model project, develops a large dataset and employs SFT and DPO to achieve performance surpassing LLaMA-2 70B and GPT-3.5 in various benchmarks and open-ended evaluations.DeepSeek872 citations56 upvotesRead →Training and scalingMay 15, 2024LoRA Learns Less and Forgets LessLoRA learns less than full fine-tuning on code and math, but forgets less of what the model already knew.Databricks486 citations91 upvotesRead →Training and scalingUnverifiedSep 16, 2024jina-embeddings-v3: Multilingual Embeddings With Task LoRAjina-embeddings-v3, a large-scale text embedding model, achieves state-of-the-art performance in multilingual and long-context retrieval tasks using Low-Rank Adaptation and Matryoshka Representation Learning.Jina AI221 citations35 upvotesRead →
About this paper
Venue
arXiv.org
Citations
6, 0 influential · Semantic Scholar
Upvotes
109 · Hugging Face
Lab
Microsoft · on Companies · on Acquisitions · on Quarterly · on Paydays · on Releases · on TechConf

Changes

CitationsNov 9, 2026Citations rose from 3 to 6.
New summaryOct 26, 2026This paper identifies termination-token mismatch as a cause of length inflation in policy distillation and provides a mitigation strategy.
TopicOct 26, 2026Moved from Foundation models to Training and scaling.
CitationsOct 12, 2026Citations rose from 2 to 3.
CitationsOct 5, 2026Citations rose from 0 to 2.
UpvotesSep 28, 2026Hugging Face upvotes rose from 107 to 109.
CitationsSep 25, 2026A first count of 0 influential citations.
CitationsSep 25, 2026A first count of 0 citations.
New paperSep 25, 2026Found by the weekly Hugging Face scan; not yet checked by hand.