An Empirical Model of Large-Batch Training
read the original abstract
In an increasing number of domains it has been demonstrated that deep learning models can be trained using relatively large batch sizes without sacrificing data efficiency. However the limits of this massive data parallelism seem to differ from domain to domain, ranging from batches of tens of thousands in ImageNet to batches of millions in RL agents that play the game Dota 2. To our knowledge there is limited conceptual understanding of why these limits to batch size differ or how we might choose the correct batch size in a new domain. In this paper, we demonstrate that a simple and easy-to-measure statistic called the gradient noise scale predicts the largest useful batch size across many domains and applications, including a number of supervised learning datasets (MNIST, SVHN, CIFAR-10, ImageNet, Billion Word), reinforcement learning domains (Atari and Dota), and even generative model training (autoencoders on SVHN). We find that the noise scale increases as the loss decreases over a training run and depends on the model size primarily through improved model performance. Our empirically-motivated theory also describes the tradeoff between compute-efficiency and time-efficiency, and provides a rough model of the benefits of adaptive batch-size training.
This paper has not been read by Pith yet.
Forward citations
Cited by 47 Pith papers
-
Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling
Pythia releases 16 identically trained LLMs with full checkpoints and data tools to study training dynamics, scaling, memorization, and bias in language models.
-
TallyTrain: Communication-Efficient Federated Distillation
TallyTrain is a hard-label distillation protocol for federated learning that uses argmax transmission and optional sparse merges to match soft-label performance at up to 1000x lower communication cost.
-
Scaling Laws for Behavioral Foundation Models over User Event Sequences
Across 600 runs from 10^15 to 10^19 FLOPs, behavioral models show a 2% embedder is compute-optimal at all scales, training is data-heavy at low compute, and optimal negatives increase with budget until memory-limited.
-
From One-Pass SGD to Data Reuse: Mini-Batch Scaling Laws in Sketched Linear Regression
Derives mini-batch scaling laws for sketched linear regression, with shared approximation terms and protocol-specific variance/fluctuation scalings under power-law spectrum and source condition.
-
Scalable Reinforcement Learning via Adaptive Batch Scaling
ABS uses Behavioral Divergence to adaptively scale batch sizes in RL according to policy volatility, enabling effective large-batch large-network training on ALE benchmarks.
-
How to Scale Mixture-of-Experts: From muP to the Maximally Scale-Stable Parameterization
The authors derive a Maximally Scale-Stable Parameterization (MSSP) for MoE models that achieves robust learning-rate transfer and monotonic performance gains with scale across co-scaling regimes of width, experts, an...
-
HEBATRON: A Hebrew-Specialized Open-Weight Mixture-of-Experts Language Model
Hebatron is the first open-weight Hebrew MoE LLM adapted from Nemotron-3, reaching 73.8% on Hebrew reasoning benchmarks while activating only 3B parameters per pass and supporting 65k-token context.
-
Hierarchical Fault Detection and Diagnosis for Transformer Architectures
DEFault++ delivers automated hierarchical fault detection, categorization into 12 transformer-specific types, and root-cause diagnosis among 45 mechanisms on a new benchmark of 3,739 mutated instances, with AUROC >0.9...
-
PACED: Distillation and On-Policy Self-Distillation at the Frontier of Student Competence
PACED applies student pass-rate weighting w(p)=p(1-p) to distillation, concentrating on the zone of proximal development and delivering up to +8.2 gains on AIME tasks with reduced forgetting.
-
How does the optimizer implicitly bias the model merging loss landscape?
Effective noise scale non-monotonically governs model merging success with an optimum, unifying effects of learning rate, weight decay, batch size, and augmentation on the loss landscape.
-
DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
DeepSeek-V2 delivers top-tier open-source LLM performance using only 21B active parameters by compressing the KV cache 93.3% and cutting training costs 42.5% via MLA and DeepSeekMoE.
-
Scaling Laws for Autoregressive Generative Modeling
Autoregressive transformers follow power-law scaling laws for cross-entropy loss with nearly universal exponents relating optimal model size to compute budget across four domains.
-
Dota 2 with Large Scale Deep Reinforcement Learning
OpenAI Five achieved superhuman performance in Dota 2 by defeating the world champions using scaled self-play reinforcement learning.
-
ZeRO: Memory Optimizations Toward Training Trillion Parameter Models
ZeRO removes memory redundancies in parallel training to scale deep learning models to over a trillion parameters with high throughput on current hardware.
-
Staleness-Learning Rate Scaling Laws for Asynchronous RLHF
Under smoothness assumptions, stale rollouts in async GRPO bias the surrogate gradient by O(S * eta) and yield the stability bound eta << min(R_batch / (S * G_upd), R_crit / (T * G_upd)).
-
On the Nonlinearity of Learning Rate Scaling for LLM Training
Optimal learning rate for models from 22M to 707M parameters shows nonlinear upward curvature with scale that disappears under effective learning rate and data-scale extrapolation.
-
Weibull Weight-Scale Parameter Evolution under AdamW Training Dynamics
Derives three-force decomposition of squared weight norm under AdamW and validates it on Pythia-70M models, plus spline recovery of alignment force from checkpoints.
-
Predictable Scaling Laws of Optimal Hyperparameters for LLM Continued Pre-training
Optimal hyperparameters for LLM continued pre-training follow predictable scaling laws derived from proxy models, enabling a two-stage framework that predicts settings from compute budget and checkpoint state to reduc...
-
Predicting Large Model Test Losses with a Noisy Quadratic System
A noisy quadratic system predicts large model test losses from N, B, K and outperforms Chinchilla's model for extrapolation up to 1000x compute.
-
Spectral Lens: Activation and Gradient Spectra as Diagnostics of LLM Optimization
Spectral analysis of activations and gradients provides new diagnostics that link batch size to representation geometry, early covariance tails to token efficiency, and spectral shifts to learning dynamics in decoder-...
-
Quantum Tilted Loss in Variational Optimization: Theory and Applications
QTL unifies expectation-value minimization with CVaR and Gibbs heuristics under one tunable operator, amplifying gradients in structured cases while preserving global minima and shifting the bottleneck to measurement ...
-
Hierarchical Fault Detection and Diagnosis for Transformer Architectures
DEFault++ applies hierarchical learning with a Fault Propagation Graph to detect, localize, and diagnose faults in transformers, improving F1 to 0.826-0.909 and developer repair accuracy from 57.1% to 83.3% on a new b...
-
COPUS: Co-adaptive Parallelism and Batch Size Selection in Large Language Model Training
COPUS co-adapts batch size and parallelism during LLM training via goodput to deliver 3.9-8% average faster convergence than fixing one while tuning the other.
-
The Recurrent Transformer: Greater Effective Depth and Efficient Decoding
Recurrent Transformers add per-layer recurrent memory via self-attention on own activations plus a tiling algorithm that reduces training memory traffic, yielding better C4 pretraining cross-entropy than parameter-mat...
-
Rethinking Language Model Scaling under Transferable Hypersphere Optimization
HyperP transfers optimal learning rates across model width, depth, tokens, and MoE granularity under Frobenius-sphere constraints, delivering stable scaling and 1.58x efficiency gains.
-
Attention Sinks Induce Gradient Sinks: Massive Activations as Gradient Regulators in Transformers
Attention sinks induce gradient sinks under causal masking, with massive activations serving as adaptive RMSNorm regulators that attenuate localized gradient pressure in Transformer training.
-
Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning
Math reasoning gains in LLMs rarely transfer to general domains; RL tuning generalizes while SFT causes forgetting and representation drift.
-
Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free
Applying a head-specific sigmoid gate after SDPA in LLMs boosts performance and stability by adding non-linearity and query-dependent sparse modulation while reducing attention sinks.
-
Language Models (Mostly) Know What They Know
Language models show good calibration when asked to estimate the probability that their own answers are correct, with performance improving as models get larger.
-
PaLM: Scaling Language Modeling with Pathways
PaLM 540B demonstrates continued scaling benefits by setting new few-shot SOTA results on hundreds of benchmarks and outperforming humans on BIG-bench.
-
A General Language Assistant as a Laboratory for Alignment
Ranked preference modeling outperforms imitation learning for language model alignment and scales more favorably with model size.
-
Scaling Laws for Transfer
Effective data transferred from pre-training to fine-tuning is described by a power law in model parameter count and fine-tuning dataset size, acting like a multiplier on the fine-tuning data.
-
How to Allocate Your Tokens? Scaling Laws with Training Steps and Batch Size
Proposes a three-term scaling law for model size, training steps and batch size that recovers optimal batch size scaling and can be fitted using fewer runs by incorporating suboptimal batch sizes.
-
Staleness-Learning Rate Scaling Laws for Asynchronous RLHF
Stale rollouts introduce O(S * eta) surrogate-gradient bias in async GRPO, yielding stability condition eta << min{R_batch / (S * G_upd), R_crit / (T * G_upd)} under smoothness assumptions.
-
Nonreversible Gauge Fields in Fokker--Planck Dynamics: Supersymmetric Hamiltonians and Learned Finite Forces
Nonreversible gauge fields are introduced for Fokker-Planck dynamics, mapped to supersymmetric Hamiltonians via similarity transformations, and optimized via finite-time actor-critic learning on Gaussian and double-we...
-
GNMR: Runtime Stability Control for Low-Precision Large Language Model Training
GNMR is a gradient-norm-based controller that maps local stability signals to budgeted recovery actions to stabilize low-precision LLM training while preserving quality.
-
The Future of Facts: Tracing the Factual Generation-Verification Gap
Empirical tracing across model families shows verification precedes and outlasts generation for facts, with updates producing simultaneous verification of old and new answers.
-
Scalable Reinforcement Learning via Adaptive Batch Scaling
Adaptive Batch Scaling dynamically increases batch size in on-policy RL as policy volatility drops, measured by a new Behavioral Divergence metric, and shows larger networks plus larger batches outperform on ALE with PQN.
-
Intelligence Inertia: Physical Isomorphism and Applications
Intelligence Inertia models the computational resistance to structural change in neural networks via a heuristic relativistic analogy, yielding a J-shaped cost curve that diverges from classical approximations.
-
GradPower: Powering Gradients for Faster Language Model Pre-Training
GradPower applies sign-power to gradients before optimization and achieves lower terminal loss in language model pre-training across architectures, scales, datasets, and schedules.
-
Parallel Manifold Steering: Efficient Adaptation of Large Associative Memories via Residual Energy Shaping
H-Res steers token trajectories via a learned vector field on the activation manifold to adapt associative memories while preserving entropy and avoiding weight changes or prompt overhead.
-
Unified Neural Scaling Laws
Presents a single functional form for neural scaling that unifies multiple scaling dimensions and claims higher extrapolation accuracy than prior forms across diverse tasks and architectures.
-
Artificial Adaptive Intelligence: The Missing Stage Between Narrow and General Intelligence
Proposes Artificial Adaptive Intelligence as the regime between narrow and general AI, defined by elimination of human-specified hyperparameters, and introduces an adaptivity index plus parametric minimality principle...
-
DeepSeek LLM: Scaling Open-Source Language Models with Longtermism
DeepSeek LLM 67B exceeds LLaMA-2 70B on code, mathematics and reasoning benchmarks after pre-training on 2 trillion tokens and alignment via SFT and DPO.
-
The Hitchhiker's Guide to Agentic AI: From Foundations to Systems
A comprehensive reference book organizing existing techniques for agentic AI systems across LLM substrate, reasoning, agent design patterns, inter-agent coordination, and production deployment.
-
There Will Be a Scientific Theory of Deep Learning
A mechanics of the learning process is emerging in deep learning theory, characterized by dynamics, coarse statistics, and falsifiable predictions across idealized settings, limits, laws, hyperparameters, and universa...
-
A Comprehensive Overview of Large Language Models
A survey paper providing an overview of Large Language Models, their background, and recent advances in the field.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.