Paper develops large time series models using pre-trained transformers.
problem Performance bottlenecks in small models on data-scarce scenarios.
method Large-scale pre-training, unified time series format, GPT-style architecture.
result Generative pre-trained Time Series Transformer (Timer) for diverse tasks.
New attack reveals memorization patterns in pre-trained LLMs.
problem Determining if a data point was part of a pre-trained LLM's training set.
method Adapts MIA statistical tests to LLM's perplexity dynamics of subsequences.
result Significantly outperforms prior approaches in membership inference attacks.
Learning-rate schedules for large models match optimization theory closely, leading to better training.
problem Improving training of large models with optimal learning rates.
method Used a bound from non-smooth convex optimization theory to match learning-rate schedules with practical benefits.
result Extending the learning-rate schedule with optimal learning-rate and transferring it across schedules improves model training.
MARS optimizes large model training by reducing variance, outperforming AdamW.
problem Training large models efficiently and scalably.
method Unified optimization framework MARS combining preconditioned gradient updates and variance reduction.
result MARS outperforms AdamW in training GPT-2 models.
AdaScale SGD adapts learning rates for large-batch training efficiently.
problem Adapting learning rates for large-batch training to balance speed-ups and model quality.
method Adaptive learning rate adaptation based on gradient variance.
result AdaScale achieves reliable speed-ups for a wide range of batch sizes without degrading model quality.
New model explains deep learning performance at large learning rates.
problem Understanding deep learning performance at different learning rates.
method Developed neural networks with solvable training dynamics.
result Large learning rates lead to convergence to flatter minima.
New method speeds up training of large kernel models.
problem Scaling kernel machines to large datasets and model sizes.
method Delayed projections in Preconditioned Stochastic Gradient Descent (PSGD).
result Significant training speed up over existing methods.
PipeDream-2BW accelerates large model training by 20x with minimal memory usage.
problem Training large models requires memory beyond single accelerator capacity.
method Pipeline parallelism, weight gradient coalescing, double buffering.
result Accelerates large model training by up to 20x.
Local SGD outperforms conventional methods in LLM training.
problem Training large language models on distributed devices.
method Local SGD algorithm applied to distributed training.
result Local SGD achieves competitive results in LLM training.
ScheduleFree+ improves large language model training without schedules or learning rates.
problem Scaling up Schedule-Free Learning to large language models.
method Learning-rate-free and schedule-free method for training large language models.
result ScheduleFree+ outperforms SOTA schedules by 31% at 1000 tokens per parameter.
Large learning rates improve neural network generalization, study shows.
problem Understanding why large learning rates lead to better neural network generalization.
method Visual analysis of training and testing loss landscapes, introduction of a nonlinear model.
result Extended phase with large learning rates leads to near-optimal generalization.
We use GG distributions to optimize LLMs, reducing size and training time.
problem Lack of understanding in the statistical structure of LLMs.
method GG-based initialization, ACT, GCT.
result Smaller, faster models with minimal communication overhead.
Many recent breakthroughs in deep learning were achieved by training increasingly larger models on massive datasets. However, training such models can be prohibitively expensive. For instance, the cluster used to train GPT-3 costs over \$250 million. As a result, most researchers cannot afford to train state of the art…
PLUMAGE improves large model training efficiency and stability.
problem Accelerator memory and networking constraints during large model training.
method Probabilistic Low rank Unbiased Minimum Variance Gradient Estimator (PLUMAGE) that resolves bias and variance issues.
result PLUMAGE reduces training loss by 28% on average across the GLUE benchmark.
Theoretical analysis of data quality and synergies in LLMs.
problem Understanding why different training methods require different amounts of data.
method Theoretical analysis of transformers trained on a weight prediction task for linear regression.
result SFT excels on smaller datasets challenging for the pretrained model, while RL benefits from large, not overly difficult data.
Adversarial training is a technique for training robust machine learning models. To encourage robustness, it iteratively computes adversarial examples for the model, and then re-trains on these examples via some update rule. This work analyzes the performance of adversarial training on linearly separable data, and prov…
SP-NGD improves deep learning models' generalization with large mini-batch sizes.
problem Worse generalization performance with large mini-batch sizes in deep learning.
method SP-NGD, a natural gradient descent approach for large-scale deep learning.
result SP-NGD achieves similar generalization performance to first-order methods with accelerated convergence and negligible overhead.
Foundation models trained on collider data improve jet generation tasks.
problem Improving foundation models for jet generation tasks.
method Pre-training OmniJet-α model on AspenOpenJets dataset. result Pre-trained model improves performance on jet generation tasks with domain shift.
Nonlinear conjugate gradient (NLCG) based optimizers have shown superior loss convergence properties compared to gradient descent based optimizers for traditional optimization problems. However, in Deep Neural Network (DNN) training, the dominant optimization algorithm of choice is still Stochastic Gradient Descent (SG…
Study builds a Japanese financial-specific LLM through continual pre-training.
problem Lack of domain-specific Japanese financial LLMs.
method Continual pre-training on Japanese financial-focused datasets using a base Japanese LLM.
result Tuned model outperforms original model on Japanese financial benchmarks.
BloombergGPT is a large language model trained on financial data, outperforming existing models on financial tasks.
problem Lack of specialized large language models for finance.
method Trained on a 363 billion token dataset augmented with 345 billion tokens from general datasets, using a 50 billion parameter model.
result BloombergGPT outperforms existing models on financial tasks without sacrificing performance on general LLM benchmarks.
Chronologically consistent models maintain accuracy with time-restricted data.
problem Training data introduces lookahead bias and training leakage in large language models.
method ChronoBERT and ChronoGPT trained with only available data at each time point.
result Models achieve strong performance and competitive with larger models, mitigating lookahead bias.
We present Distributed Equivalent Substitution (DES) training, a novel distributed training framework for large-scale recommender systems with dynamic sparse features. DES introduces fully synchronous training to large-scale recommendation system for the first time by reducing communication, thus making the training of…
Stochastic gradient descent with a large initial learning rate is widely used for training modern neural net architectures. Although a small initial learning rate allows for faster training and better test performance initially, the large learning rate achieves better generalization soon after the learning rate is anne…
Model collapse occurs even with minimal synthetic data, and larger models can exacerbate it.
problem Model collapse in large neural networks due to synthetic data.
method Theoretical and empirical investigation of model collapse in various neural network architectures.
result Model collapse is exacerbated by larger models, but not entirely prevented.
We introduce a scalable measure of curvature for analyzing training dynamics of large language models.
problem Analyzing the training dynamics of large language models due to high computational cost of measuring Hessian sharpness.
method We introduce critical sharpness and relative critical sharpness as computationally efficient measures capturing Hessian sharpness phenomena.
result We provide the first demonstration of sharpness phenomena at scale up to 7B parameters.
We study the problem of learning similarity functions over very large corpora using neural network embedding models. These models are typically trained using SGD with sampling of random observed and unobserved pairs, with a number of samples that grows quadratically with the corpus size, making it expensive to scale to…
LIIT uses large deviations to train neural networks faster.
problem Training neural networks requires large amounts of data, leading to inefficiencies.
method LIIT uses large deviations principle to generate and update training samples.
result LIIT achieves similar classification performance with smaller, more representative samples.
We propose Stochastic Weight Averaging in Parallel (SWAP), an algorithm to accelerate DNN training. Our algorithm uses large mini-batches to compute an approximate solution quickly and then refines it by averaging the weights of multiple models computed independently and in parallel. The resulting models generalize equ…
Deep Neural Networks (DNNs) often rely on very large datasets for training. Given the large size of such datasets, it is conceivable that they contain certain samples that either do not contribute or negatively impact the DNN's optimization. Modifying the training distribution in a way that excludes such samples could …
New framework trains large SciML models solving PDEs in reasonable time.
problem Training large SciML models solving PDEs is challenging and time-consuming.
method Data parallel distributed deep learning framework with optimized methods.
result Neural PDE solvers can be viably trained for practical applications.
Training convolutional networks (CNN's) that fit on a single GPU with minibatch stochastic gradient descent has become effective in practice. However, there is still no effective method for training large CNN's that do not fit in the memory of a few GPU cards, or for parallelizing CNN training. In this work we show tha…
New framework reveals thermodynamic principles for LLM training.
problem Understanding the training dynamics of large language models.
method Introducing Neural Thermodynamic Laws (NTL) under river-valley loss landscape assumptions.
result Key thermodynamic quantities and principles naturally emerge in LLM training.
TFB simplifies Bayesian LLM uncertainty estimation without extra training.
problem Estimating uncertainty in LLM responses remains challenging.
method Training-Free Bayesianization (TFB) that transforms low-rank adapters into Bayesian ones without additional training.
result TFB achieves superior uncertainty estimation and generalization compared to existing methods.
Recently, pre-trained language representation flourishes as the mainstay of the natural language understanding community, e.g., BERT. These pre-trained language representations can create state-of-the-art results on a wide range of downstream tasks. Along with continuous significant performance improvement, the size an…
SNGM improves large-batch training accuracy.
problem Improving generalization in large-batch training.
method Stochastic Normalized Gradient Descent with Momentum.
result SNGM achieves better test accuracy than MSGD and other large-batch methods.
Spectrum selectively trains LLMs based on SNR to save resources.
problem Efficiently training large language models with limited resources.
method Targeting layer modules based on SNR for selective training.
result Spectrum achieves similar performance to full fine-tuning but with reduced VRAM usage.
Survey on using large models to train smaller datasets in NLP.
problem Lack of large datasets and computing resources for NLP tasks.
method Analysis of recent transfer learning approaches in NLP.
result Increased demand for transfer learning in NLP due to large models.
Optimus pre-trains sentences in a latent space for various NLP tasks.
problem Training large-scale language models for diverse NLP tasks.
method Pre-trained Variational Autoencoder (VAE) on large text corpus, fine-tuned for various tasks.
result Optimus achieves state-of-the-art on VAE language modeling benchmarks.
Derives ideal train/test split for ridge regression in large data limit.
problem Finding optimal train/test split for ridge regression in large data scenarios.
method Mathematical derivation of optimal train/test split, considering ridge tuning parameter and asymptotic behavior.
result The optimal train/test split for ridge regression in the large data limit depends weakly on the ridge tuning parameter alpha.
Law explains how LLMs learn to predict next tokens.
problem Understanding how LLMs process input data internally.
method Introduced a precise law governing token embeddings in LLMs.
result Each layer equally contributes to next-token prediction accuracy.
Many of the recent approaches to polyphonic piano note onset transcription require training a machine learning model on a large piano database. However, such approaches are limited by dataset availability; additional training data is difficult to produce, and proposed systems often perform poorly on novel recording con…
Large models follow power laws in performance with dataset size or parameters.
problem Understanding neural scaling laws in large language models.
method Joint generative data model and random feature model.
result Modeling and solving the dual limit reveals insights into scaling laws.
The paper simplifies influence computations for large-scale machine learning models.
problem Improving training efficiency and accuracy in large-scale models.
method Study influence functions, define memorization, simplify computations.
result Influence functions can be practical for large-scale models, indicating memorization.
Paper proposes LMM-PQS for cross-domain few-shot learning.
problem Cross-domain few-shot learning problem.
method Generates pseudo query images and fine-tunes feature extraction modules with a large margin mechanism.
result LMM-PQS outperforms baseline models in cross-domain few-shot learning.
Extends hyperparameter transfer across model sizes and modules, improving training speed.
problem Training stability and performance of large-scale models with optimal hyperparameters.
method Complete(d) Parameterisation, per-module hyperparameter optimisation and transfer. result Hyperparameter transfer holds even in the per-module hyperparameter regime, improving training speed.
We determine the critical batch size for large language models and find it scales with data size, not model size.
problem Determining the optimal batch size for large-scale model training.
method We propose a measure of critical batch size, pre-trained models, and systematic hyper-parameter sweeps.
result The critical batch size scales primarily with data size, not model size.
Efficiently trains deep Gaussian processes on large datasets.
problem Large-scale data and multi-scale features in function approximation.
method Combines variational learning with MCMC for efficient and accurate training.
result Highly efficient and accurate deep GP training on large-scale data.