Pith. sign in

REVIEW 12 cited by

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2304.09871 v2 pith:SO76PEDC submitted 2023-04-19 cs.LG cs.AImath.OC

A Theory on Adam Instability in Large-Scale Machine Learning

classification cs.LG cs.AImath.OC
keywords trainingbillionadamlanguagelargetheoryargueartifact
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

We present a theory for the previously unexplained divergent behavior noticed in the training of large language models. We argue that the phenomenon is an artifact of the dominant optimization algorithm used for training, called Adam. We observe that Adam can enter a state in which the parameter update vector has a relatively large norm and is essentially uncorrelated with the direction of descent on the training loss landscape, leading to divergence. This artifact is more likely to be observed in the training of a deep model with a large batch size, which is the typical setting of large-scale language model training. To argue the theory, we present observations from the training runs of the language models of different scales: 7 billion, 30 billion, 65 billion, and 546 billion parameters.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 12 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Grokking or Glitching? How Low-Precision Drives Slingshot Loss Spikes

    cs.LG 2026-05 unverdicted novelty 7.0

    Slingshot loss spikes are produced by low-precision arithmetic that breaks the zero-sum gradient constraint and drives exponential growth via Numerical Feature Inflation.

  2. Grokking or Glitching? How Low-Precision Drives Slingshot Loss Spikes

    cs.LG 2026-05 unverdicted novelty 7.0

    Slingshot loss spikes arise from floating-point precision limits that round correct-class gradients to zero, breaking zero-sum constraints and driving exponential parameter growth through numerical feature inflation.

  3. Grokking or Glitching? How Low-Precision Drives Slingshot Loss Spikes

    cs.LG 2026-05 unverdicted novelty 7.0

    Slingshot loss spikes result from floating-point precision limits that round correct-class gradients to zero, triggering Numerical Feature Inflation and breaking gradient zero-sum constraints.

  4. Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention

    cs.LG 2025-10 unverdicted novelty 7.0

    Low-precision Flash Attention fails due to similar low-rank attention representations combined with biased rounding errors that accumulate and corrupt weight updates; a minimal fix to reduce rounding bias stabilizes training.

  5. SCAPE: Accurate and Efficient LLM Training with Extreme Sparse Communication

    cs.LG 2026-07 conditional novelty 6.0

    SCAPE enables 90-99% sparse gradient communication in sharded Adam-style LLM training by deriving masks from first-moment statistics, achieving up to 43.3% faster pre-training on Llama-500M with no loss in validation ...

  6. MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention

    cs.CL 2025-06 unverdicted novelty 6.0

    MiniMax-M1 is a 456B parameter hybrid-attention MoE model trained with CISPO RL that achieves performance comparable or superior to DeepSeek-R1 and Qwen3-235B on reasoning and software engineering tasks while training...

  7. GWT: Scalable Optimizer State Compression for Large Language Model Training

    cs.LG 2025-01 unverdicted novelty 6.0

    GWT projects gradients into wavelet subspaces to compress optimizer states for memory-efficient LLM training while claiming performance parity with full-rank updates.

  8. Emerging Properties in Unified Multimodal Pretraining

    cs.CV 2025-05 unverdicted novelty 5.0

    BAGEL is a unified decoder-only model that develops emerging complex multimodal reasoning abilities after pretraining on large-scale interleaved data and outperforms prior open-source unified models.

  9. Open-Sora: Democratizing Efficient Video Production for All

    cs.CV 2024-12 unverdicted novelty 5.0

    Open-Sora releases an open-source video generation model based on a Spatial-Temporal Diffusion Transformer that decouples spatial and temporal attention, supporting text-to-video, image-to-video, and text-to-image tas...

  10. PowLU: An Activation Function for Stable Pre-Training of LLMs

    cs.CL 2026-05 unverdicted novelty 4.0

    PowLU replaces SwiGLU with a rational-power activation to reduce outlier amplification and numerical instability during large-scale LLM pre-training while matching performance.

  11. Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

    cs.CL 2025-07 unverdicted novelty 4.0

    Gemini 2.5 Pro and Flash models are presented as achieving frontier performance in reasoning, coding, and long-context multimodal tasks while spanning a cost-capability Pareto curve.

  12. Navigating LLM Valley: From AdamW to Memory-Efficient and Matrix-Based Optimizers

    cs.LG 2026-05 unverdicted novelty 3.0

    This survey organizes LLM optimizer literature into categories and argues the field is shifting toward rigorous, multi-factor comparisons of convergence, memory, stability, and complexity.