REVIEW 9 cited by
Resolving Discrepancies in Compute-Optimal Scaling of Language Models
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
Resolving Discrepancies in Compute-Optimal Scaling of Language Models
read the original abstract
Kaplan et al. and Hoffmann et al. developed influential scaling laws for the optimal model size as a function of the compute budget, but these laws yield substantially different predictions. We explain the discrepancy by reproducing the Kaplan scaling law on two datasets (OpenWebText2 and RefinedWeb) and identifying three factors causing the difference: last layer computational cost, warmup duration, and scale-dependent optimizer tuning. With these factors corrected, we obtain excellent agreement with the Hoffmann et al. (i.e., "Chinchilla") scaling law. Counter to a hypothesis of Hoffmann et al., we find that careful learning rate decay is not essential for the validity of their scaling law. As a secondary result, we derive scaling laws for the optimal learning rate and batch size, finding that tuning the AdamW $\beta_2$ parameter is essential at lower batch sizes.
Forward citations
Cited by 9 Pith papers
-
How Much Is a Dataset Worth? Scaling Laws, the Vendi Score, and Matrix Spectral Functions
Vendi Score and scaling-law objectives belong to the class of matrix spectral functions, which are submodular, enabling efficient greedy selection of training data that outperforms random subsets in predicting held-ou...
-
Scaling Laws for Mixture Pretraining Under Data Constraints
Repetition-aware scaling laws show scarce target data in pretraining mixtures can be repeated 15-20 times optimally, with the best count depending on data size, compute, and model scale.
-
Deriving Neural Scaling Laws from the statistics of natural language
The data-limited loss exponent of LLMs equals γ/(2β), where γ is the context-length decay of next-token entropy and β is the lag decay of token correlations, matching experiments on two corpora.
-
The Art of Scaling Reinforcement Learning Compute for LLMs
A 400k+ GPU-hour study shows RL scaling in LLMs follows predictable sigmoidal trajectories, with most design choices affecting efficiency rather than the performance asymptote, enabling accurate large-scale prediction...
-
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.
-
How Much Is a Dataset Worth? Scaling Laws, the Vendi Score, and Matrix Spectral Functions
Facility location — a classic submodular coverage objective — predicts a training subset's held-out accuracy far better than the Vendi score, which becomes misleading at high values.
-
Mix, Don't Tune: Bilingual Pre-Training Outperforms Hyperparameter Search in Data-Constrained Settings
Mixing auxiliary high-resource language data outperforms hyperparameter tuning in data-constrained bilingual pre-training, with gains equivalent to 2-13 times more unique target data.
-
Scaling Laws for Mixture Pretraining Under Data Constraints
Empirical study shows mixture pretraining tolerates higher target data repetition than single-source training, with a new repetition-aware scaling law enabling principled mixture selection based on data size, compute,...
-
LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws
Pretraining data determines loss-to-loss scaling laws in LLMs, while model size, optimization, tokenizer, and architecture have limited impact.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.