ML Cult

Think before you speak: Training Language Models With Pause TokensTowards Self-Assembling Artificial Neural Networks through Neural Developmental ProgramsEfficient Streaming Language Models with Attention SinksLarge Language Models Cannot Self-Correct Reasoning YetSmartPlay : A Benchmark for LLMs as Intelligent AgentsSupport the show

10-04

13:09

Enable Language Models to Implicitly Learn Self-Improvement From DataPixArt-alpha: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image SynthesisFELM: Benchmarking Factuality Evaluation of Large Language ModelsSupport the show

10-03

07:51

ToRA: A Tool-Integrated Reasoning Agent for Mathematical Problem SolvingDecaf: Monocular Deformation Capture for Face and Hand InteractionsDirectly Fine-Tuning Diffusion Models on Differentiable RewardsSupport the show

10-02

06:52

Effective Long-Context Scaling of Foundation ModelsDemystifying CLIP DataVision Transformers Need RegistersQwen Technical ReportDreamGaussian: Generative Gaussian Splatting for Efficient 3D Content CreationSupport the show

09-29

16:14

NeuRBF: A Neural Fields Representation with Adaptive Radial Basis FunctionsEmu: Enhancing Image Generation Models Using Photogenic Needles in a HaystackShow-1: Marrying Pixel and Latent Diffusion Models for Text-to-Video GenerationFinite Scalar Quantization: VQ-VAE Made SimpleSupport the show

09-28

08:49

DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer ModelsAligning Large Multimodal Models with Factually Augmented RLHFLAVIE: High-Quality Video Generation with Cascaded Latent Diffusion ModelsSupport the show

09-27

06:30

CoRF : Colorizing Radiance Fields using Knowledge DistillationThe Cambridge Law Corpus: A Corpus for Legal AI ResearchCodePlan: Repository-level Coding using LLMs and PlanningDualToken-ViT: Position-aware Efficient Vision Transformer with Dual Token FusionSupport the show

09-25

10:47

Parallelizing non-linear sequential models over the sequence lengthFast Feedforward NetworksLongLoRA: Efficient Fine-tuning of Long-Context Large Language ModelsA Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language ModelsBoolformer: Symbolic Regression of Logic Functions with TransformersSupport the show

09-22

13:54

FreeU: Free Lunch in Diffusion U-NetNeurons in Large Language Models: Dead, N-gram, PositionalDreamLLM: Synergistic Multimodal Comprehension and CreationKosmos-2.5: A Multimodal Literate ModelEnd-to-End Speech Recognition Contextualization with Large Language ModelsThe Languini Kitchen: Enabling Language Modelling Research at Different Scales of ComputeSupport the show

09-21

14:32

Graph Neural Networks Use Graphs When They Shouldn'tLarge Language Models for Compiler OptimizationOpenBA: An Open-sourced 15B Bilingual Asymmetric seq2seq Model Pre-trained from ScratchBaichuan 2: Open Large-scale Language ModelsLanguage Modeling Is CompressionFoleyGen: Visually-Guided Audio GenerationSupport the show

09-20

13:20

Textbooks Are All You Need II: phi-1.5 technical reportDiffBIR: Towards Blind Image Restoration with Generative Diffusion PriorWhen Less is More: Investigating Data Pruning for Pretraining LLMs at ScaleMADLAD-400: A Multilingual And Document-Level Large Audited DatasetFIAT: Fusing learning paradigms with Instruction-Accelerated TuningOptimize Weight Rounding via Signed Gradient Descent for the Quantization of LLMsSupport the show

09-12

11:11

Large-Scale Automatic Audiobook CreationCityDreamer: Compositional Generative Model of Unbounded 3D CitiesFrom Sparse to Dense: GPT-4 Summarization with Chain of Density PromptingMobile V-MoEs: Scaling Down Vision Transformers via Sparse Mixture-of-ExpertsHigh-Quality Entity SegmentationSupport the show

09-11

09:26

Large Language Models as OptimizersFLM-101B: An Open LLM and How to Train It with $100K BudgetXGen-7B Technical ReportTracking Anything with Decoupled Video SegmentationDoLa: Decoding by Contrasting Layers Improves Factuality in Large Language ModelsSupport the show

09-08

11:28

SLiMe: Segment Like MeMatcha-TTS: A fast TTS architecture with conditional flow matchingPhysically Grounded Vision-Language Models for Robotic ManipulationScaling Autoregressive Multi-Modal Models: Pretraining and Instruction TuningSupport the show

09-07

08:11

One Wide Feedforward is All You NeedEfficient RLHF: Reducing the Memory Usage of PPOPromptTTS 2: Describing and Generating Voices with Text PromptAniPortraitGAN: Animatable 3D Portrait Generation from 2D Image CollectionsSupport the show