The Impact of Depth and Width on Transformer Language Model Generalization
Paper
•
2310.19956
•
Published
•
9
LLM architecture
-
-
Retentive Network: A Successor to Transformer for Large Language Models
Paper • 2307.08621 • Published • 170 -
RWKV: Reinventing RNNs for the Transformer Era
Paper • 2305.13048 • Published • 14 -
Attention Is All You Need
Paper • 1706.03762 • Published • 44 -
READ: Recurrent Adaptation of Large Transformers
Paper • 2305.15348 • Published • 2 -
Emergent Mixture-of-Experts: Can Dense Pre-trained Transformers Benefit from Emergent Modular Structures?
Paper • 2310.10908 • Published • 1 -
Parameter-Efficient Mixture-of-Experts Architecture for Pre-trained Language Models
Paper • 2203.01104 • Published • 2 -
Scaling Pre-trained Language Models to Deeper via Parameter-efficient Architecture
Paper • 2303.16753 • Published • 1 -
GateLoop: Fully Data-Controlled Linear Recurrence for Sequence Modeling
Paper • 2311.01927 • Published • 1 -
White-Box Transformers via Sparse Rate Reduction
Paper • 2306.01129 • Published • 1 -
Improving Transformers with Probabilistic Attention Keys
Paper • 2110.08678 • Published • 1 -
Wide Attention Is The Way Forward For Transformers?
Paper • 2210.00640 • Published • 1 -
Architecture Matters in Continual Learning
Paper • 2202.00275 • Published • 1 -
Scaling TransNormer to 175 Billion Parameters
Paper • 2307.14995 • Published • 21 -
FlashFFTConv: Efficient Convolutions for Long Sequences with Tensor Cores
Paper • 2311.05908 • Published • 12 -
Hiformer: Heterogeneous Feature Interactions Learning with Transformers for Recommender Systems
Paper • 2311.05884 • Published • 5 -
AutoML in the Age of Large Language Models: Current Challenges, Future Opportunities and Risks
Paper • 2306.08107 • Published • 1 -
Continual Learning with Dependency Preserving Hypernetworks
Paper • 2209.07712 • Published • 1 -
Are We Falling in a Middle-Intelligence Trap? An Analysis and Mitigation of the Reversal Curse
Paper • 2311.07468 • Published • 1 -
Rethinking Attention: Exploring Shallow Feed-Forward Neural Networks as an Alternative to Attention Layers in Transformers
Paper • 2311.10642 • Published • 23 -
Trellis Networks for Sequence Modeling
Paper • 1810.06682 • Published • 1 -
ProSG: Using Prompt Synthetic Gradients to Alleviate Prompt Forgetting of RNN-like Language Models
Paper • 2311.01981 • Published • 1 -
Exponentially Faster Language Modelling
Paper • 2311.10770 • Published • 118 -
Replacing softmax with ReLU in Vision Transformers
Paper • 2309.08586 • Published • 17 -
Transformer Language Models without Positional Encodings Still Learn Positional Information
Paper • 2203.16634 • Published • 5 -
Maestro: Uncovering Low-Rank Structures via Trainable Decomposition
Paper • 2308.14929 • Published • 1 -
Robust low-rank training via approximate orthonormal constraints
Paper • 2306.01485 • Published • 1 -
Low Rank Optimization for Efficient Deep Learning: Making A Balance between Compact Architecture and Fast Training
Paper • 2303.13635 • Published • 1 -
Cuttlefish: Low-Rank Model Training without All the Tuning
Paper • 2305.02538 • Published • 1 -
Relaxed Attention for Transformer Models
Paper • 2209.09735 • Published • 1 -
I3D: Transformer architectures with input-dependent dynamic depth for speech recognition
Paper • 2303.07624 • Published • 1 -
Emergence of Segmentation with Minimalistic White-Box Transformers
Paper • 2308.16271 • Published • 13 -
White-Box Transformers via Sparse Rate Reduction: Compression Is All There Is?
Paper • 2311.13110 • Published • 1 -
Linear Self-Attention Approximation via Trainable Feedforward Kernel
Paper • 2211.04076 • Published • 1 -
Mamba: Linear-Time Sequence Modeling with Selective State Spaces
Paper • 2312.00752 • Published • 138 -
Decoder-only Architecture for Speech Recognition with CTC Prompts and Text Data Augmentation
Paper • 2309.08876 • Published • 1 -
Smooth Diffusion: Crafting Smooth Latent Spaces in Diffusion Models
Paper • 2312.04410 • Published • 14 -
HyperMixer: An MLP-based Low Cost Alternative to Transformers
Paper • 2203.03691 • Published • 1 -
Interpret Vision Transformers as ConvNets with Dynamic Convolutions
Paper • 2309.10713 • Published • 1 -
Laughing Hyena Distillery: Extracting Compact Recurrences From Convolutions
Paper • 2310.18780 • Published • 3 -
LLM360: Towards Fully Transparent Open-Source LLMs
Paper • 2312.06550 • Published • 57 -
Text Generation with Diffusion Language Models: A Pre-training Approach with Continuous Paragraph Denoise
Paper • 2212.11685 • Published • 2 -
AR-Diffusion: Auto-Regressive Diffusion Model for Text Generation
Paper • 2305.09515 • Published • 2 -
TESS: Text-to-Text Self-Conditioned Simplex Diffusion
Paper • 2305.08379 • Published • 1 -
DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models
Paper • 2210.08933 • Published • 5 -
DiffuSIA: A Spiral Interaction Architecture for Encoder-Decoder Text Diffusion
Paper • 2305.11517 • Published • 1 -
Diffusion Language Models Can Perform Many Tasks with Scaling and Instruction-Finetuning
Paper • 2308.12219 • Published • 1 -
Likelihood-Based Diffusion Language Models
Paper • 2305.18619 • Published • 1 -
ParaGuide: Guided Diffusion Paraphrasers for Plug-and-Play Textual Style Transfer
Paper • 2308.15459 • Published • 1 -
SSD-LM: Semi-autoregressive Simplex-based Diffusion Language Model for Text Generation and Modular Control
Paper • 2210.17432 • Published • 1 -
Self-conditioned Embedding Diffusion for Text Generation
Paper • 2211.04236 • Published • 1 -
Cached Transformers: Improving Transformers with Differentiable Memory Cache
Paper • 2312.12742 • Published • 12 -
SwitchGPT: Adapting Large Language Models for Non-Text Outputs
Paper • 2309.07623 • Published • 1 -
Learning to Skip for Language Modeling
Paper • 2311.15436 • Published • 1 -
Zoology: Measuring and Improving Recall in Efficient Language Models
Paper • 2312.04927 • Published • 2 -
Beyond Surface: Probing LLaMA Across Scales and Layers
Paper • 2312.04333 • Published • 18 -
DeLighT: Deep and Light-weight Transformer
Paper • 2008.00623 • Published • 1 -
Leveraging Contextual Information for Effective Entity Salience Detection
Paper • 2309.07990 • Published • 7 -
Block-State Transformer
Paper • 2306.09539 • Published • 9 -
Blockwise Parallel Transformer for Long Context Large Models
Paper • 2305.19370 • Published • 3 -
Block-Recurrent Transformers
Paper • 2203.07852 • Published • 1 -
Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model
Paper • 2401.09417 • Published • 59 -
RNNs of RNNs: Recursive Construction of Stable Assemblies of Recurrent Neural Networks
Paper • 2106.08928 • Published • 1 -
LKCA: Large Kernel Convolutional Attention
Paper • 2401.05738 • Published • 1 -
InfoDiffusion: Information Entropy Aware Diffusion Process for Non-Autoregressive Text Generation
Paper • 2310.11976 • Published • 2 -
Enhancing Phrase Representation by Information Bottleneck Guided Text Diffusion Process for Keyphrase Extraction
Paper • 2308.08739 • Published • 1 -
Gated Linear Attention Transformers with Hardware-Efficient Training
Paper • 2312.06635 • Published • 6 -
Hierarchically Gated Recurrent Neural Network for Sequence Modeling
Paper • 2311.04823 • Published • 2 -
Improving Natural Language Capability of Code Large Language Model
Paper • 2401.14242 • Published • 1 -
BlackMamba: Mixture of Experts for State-Space Models
Paper • 2402.01771 • Published • 23 -
Can Mamba Learn How to Learn? A Comparative Study on In-Context Learning Tasks
Paper • 2402.04248 • Published • 30 -
Simple Hardware-Efficient Long Convolutions for Sequence Modeling
Paper • 2302.06646 • Published • 2 -
A Quantitative Review on Language Model Efficiency Research
Paper • 2306.01768 • Published • 1 -
A Unified View of Long-Sequence Models towards Modeling Million-Scale Dependencies
Paper • 2302.06218 • Published • 1 -
Accelerating Toeplitz Neural Network with Constant-time Inference Complexity
Paper • 2311.08756 • Published • 1 -
Agent Attention: On the Integration of Softmax and Linear Attention
Paper • 2312.08874 • Published • 2 -
Linear Transformers with Learnable Kernel Functions are Better In-Context Models
Paper • 2402.10644 • Published • 79 -
Enhancing Transformer RNNs with Multiple Temporal Perspectives
Paper • 2402.02625 • Published -
ByT5: Towards a token-free future with pre-trained byte-to-byte models
Paper • 2105.13626 • Published • 2 -
Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Models
Paper • 2402.19427 • Published • 52 -
Simple linear attention language models balance the recall-throughput tradeoff
Paper • 2402.18668 • Published • 18 -
Linear Transformers are Versatile In-Context Learners
Paper • 2402.14180 • Published • 6 -
DenseMamba: State Space Models with Dense Hidden Connection for Efficient Large Language Models
Paper • 2403.00818 • Published • 15 -
MEGABYTE: Predicting Million-byte Sequences with Multiscale Transformers
Paper • 2305.07185 • Published • 9 -
OpenELM: An Efficient Language Model Family with Open-source Training and Inference Framework
Paper • 2404.14619 • Published • 126 -
Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length
Paper • 2404.08801 • Published • 63 -
Various Lengths, Constant Speed: Efficient Language Modeling with Lightning Attention
Paper • 2405.17381 • Published -
The Hedgehog & the Porcupine: Expressive Linear Attentions with Softmax Mimicry
Paper • 2402.04347 • Published • 13 -
SVD-LLM: Truncation-aware Singular Value Decomposition for Large Language Model Compression
Paper • 2403.07378 • Published • 3 -
GoldFinch: High Performance RWKV/Transformer Hybrid with Linear Pre-Fill and Extreme KV-Cache Compression
Paper • 2407.12077 • Published • 54