Trinomial Scaling Law Jointly Derives Optimal Batch Size
orvieto_antonio · x · 2026-07-04
Addressing the compute allocation trade-off between more steps and larger batches, the author proposes a trinomial loss law L(N,M,K)= E + A·N⁻ᵅ + B·M⁻ᵝ + C·K⁻ᵞ, where M is the token-based batch size and K is the number of training steps. Unlike previous approaches that fit batch-size laws separately, the batch size here directly factors into the loss law, allowing optimal and critical batch sizes to be derived rather than fitted independently. Since D=MK (total tokens), the law predicts an optimal batch size M ∝ D^(γ/(β+γ)). A link to the paper is included.
More from Research
- Researchers identify "Forking": train loss falls while validation loss turns up — ZimingLiu11 · 2026-10-26
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
- Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26