data-scaling-outweighs-parameter-scaling

IN derived (depth 1)

Created 2026-06-21T09:52:14+00:00 · Reviewed 2026-06-21T14:41:08+00:00

Empirical results consistently show data volume matters more than parameter count: Chinchilla demonstrated models were undertrained, Llama 1 13B beat GPT-3 175B, and Llama 3 8B continued improving at 75x Chinchilla-optimal data.

Summary

Across multiple independent experiments, the pattern is clear: feeding a smaller model far more training data consistently beats training a much larger model on less data. This means the most cost-effective path to better LLM performance is investing in more high-quality training data rather than simply building a bigger network.

Justifications

SL — Three independent results (scaling law theory, cross-model benchmark, continued improvement past optimal) all point to data as the more important scaling axis

Antecedents (all must be IN):

  • IN chinchilla-scaling-balance — The Chinchilla-style insight (Hoffmann et al., 2022) showed that LLM performance depends not on parameter count alone but on how model size, data, and compute are balanced during training.
  • IN llama1-13b-outperformed-gpt3-175b — Llama 1 13B outperformed GPT-3 175B on most NLP benchmarks, demonstrating the value of data scaling over parameter scaling
  • IN llama3-8b-trained-15t-tokens-chinchilla-suboptimal — Llama 3 8B was trained on 15T tokens — 75x more than the Chinchilla-optimal 200B tokens — and performance continued to scale log-linearly, challenging Chinchilla scaling assumptions

Dependents

These beliefs depend on this one: