āĻĒā§āϰāϝ⧁āĻ•ā§āϤāĻŋāĻ—āϤ āĻ—āĻžāχāĻĄ

Data-Constrained Scaling Laws

Data-constrained scaling studies how to allocate training resources when the supply of usable unique data is limited.

āĻāχ āύāĻŋāĻ°ā§āĻĻ⧇āĻļāĻŋāĻ•āĻž āĻļ⧁āύ⧁āύāĻāφāχ-āĻŦāĻ°ā§āĻŖāĻŋāϤ ¡ ~3 min

āĻāχ āĻĒ⧃āĻˇā§āĻ āĻžāϝāĻŧ3 āĻŽāĻŋāύāĻŋāϟ āĻĒāĻĄāĻŧāĻž āĻšāϝāĻŧ⧇āϛ⧇
  1. āĻ“āĻ­āĻžāϰāĻ­āĻŋāω
  2. āĻ—āĻ­ā§€āϰ āĻĄā§āĻŦ
  3. āĻŦāĻžāĻ¸ā§āϤāĻŦ-āĻŦāĻŋāĻļā§āĻŦ āĻŦāĻžāĻ¸ā§āϤāĻŦāĻžāϝāĻŧāύ
  4. The Future of Data-Constrained Scaling Laws
  5. āĻāϟāĻžāϕ⧇ āĻŦāĻžāĻ¸ā§āϤāĻŦ⧇ āĻĒā§āϰāϝāĻŧā§‹āĻ— āĻ•āϰāĻž āĻšāĻšā§āϛ⧇
  6. āĻ…āĻ¨ā§āĻŦ⧇āώāĻŖ āϚāĻžāϞāĻŋāϝāĻŧ⧇ āϝāĻžāύ
  7. āĻĒā§āϰāĻžāϝāĻŧāĻļāχ āϜāĻŋāĻœā§āĻžāĻžāϏāĻŋāϤ āĻĒā§āϰāĻļā§āύāĻžāĻŦāϞ⧀

āĻ“āĻ­āĻžāϰāĻ­āĻŋāω

Repeating a corpus increases processed tokens without creating the same amount of new information. A useful plan distinguishes unique tokens, repeated exposures, model size, and evidence from held-out evaluation.

āĻ—āĻ­ā§€āϰ āĻĄā§āĻŦ

A training budget may permit more computation than the available corpus can use efficiently in a single pass. This happens in small domains, restricted language collections, and projects with a fixed set of approved documents. The question becomes how much to repeat the data, how large a model to train, and whether acquiring or improving data is more valuable than extending the run. Keep two token counts. Unique tokens describe the available corpus after the chosen processing and duplicate treatment. Processed tokens count every training exposure. With ten billion unique tokens and three complete passes, the run processes thirty billion tokens. The arithmetic tracks work, not thirty billion distinct pieces of evidence. A tokenizer or filtering change can also change the token count, so preserve the counting method when comparing experiments. The study Scaling Data-Constrained Language Models explicitly investigates this setting. Its experiments found that a limited amount of repetition could approach the loss achieved with additional unique data, while further repetition eventually offered diminishing value. The reported finding of small loss differences through roughly four epochs belongs to that study’s tested conditions; it is not a universal instruction to train every corpus four times. The work proposes a fitted scaling model that accounts for the reduced value of repeated data. Use that result to design a local experiment, not to skip one. Compare model sizes and repetition schedules under a stated compute budget, retain held-out data, and measure relevant task performance alongside loss. Keep evaluation examples out of training and check duplicate overlap. If later checkpoints improve training loss but worsen validation results, investigate memorization or mismatch rather than assuming more compute must help. Consider data quality, coverage, and permission to use candidate additions before enlarging the corpus.

āĻŦāĻžāĻ¸ā§āϤāĻŦ-āĻŦāĻŋāĻļā§āĻŦ āĻŦāĻžāĻ¸ā§āϤāĻŦāĻžāϝāĻŧāύ

A hypothetical team has ten billion unique tokens and runs three complete epochs. It records thirty billion processed tokens, while the unique-data count remains ten billion.

Two experiments process the same total tokens: one uses more unique documents and the other repeats a smaller corpus. The team compares held-out loss instead of assuming the token totals imply equal learning.

A group adds code to a text corpus as a candidate data mixture, then checks both text tasks and code tasks rather than presuming that every added token benefits every use case.

A researcher observes training loss falling while held-out loss rises after more repetitions. The team keeps the earlier checkpoint and investigates before spending the remaining compute.

The Future of Data-Constrained Scaling Laws

As organizations train on specialized and curated collections, repetition schedules and data mixtures will remain practical design choices. Better fitted models may help prioritize experiments, but their predictions should be checked when the domain, architecture, tokenizer, or quality filters change. Record unique and processed tokens separately, preserve training and evaluation provenance, and compare checkpoints at meaningful budget intervals. An improved forecast is useful when it helps avoid an unproductive run; it does not remove the need to measure whether the resulting model performs the intended tasks.

āĻāϟāĻžāϕ⧇ āĻŦāĻžāĻ¸ā§āϤāĻŦ⧇ āĻĒā§āϰāϝāĻŧā§‹āĻ— āĻ•āϰāĻž āĻšāĻšā§āϛ⧇

āϕ⧇āύ āĻāϟāĻž āϗ⧁āϰ⧁āĻ¤ā§āĻŦāĻĒā§‚āĻ°ā§āĻŖ

  • āĻ–āϰāϚ āĻ“ āĻŦāĻžāĻœā§‡āϟ. āφāĻ°ā§āĻ•āĻŋāĻŸā§‡āĻ•āϚāĻžāϰ⧇āϰ āϏāĻŋāĻĻā§āϧāĻžāĻ¨ā§āϤāϗ⧁āϞāĻŋ āĻŦāĻ›āϰ⧇āϰ āĻĒāϰ āĻŦāĻ›āϰ āϧāϰ⧇ āĻ•āĻ°ā§āĻŽāĻ•ā§āώāĻŽāϤāĻž āĻāĻŦāĻ‚ āĻ…āĻĒāĻžāϰ⧇āϟāĻŋāĻ‚ āĻ–āϰāϚ āϚāĻžāϞāĻžāϝāĻŧāĨ¤
  • āϏ⧁āĻ¸ā§āĻĒāĻˇā§āϟ āϏāĻŋāĻĻā§āϧāĻžāĻ¨ā§āϤ. āĻ•āĻžāϰāĻŋāĻ—āϰāĻŋ āĻļāĻŋāĻ•ā§āώāĻž āĻĻāϞāϗ⧁āϞ⧋āϕ⧇ āϏāĻ āĻŋāĻ• āĻ¸ā§āĻŸā§āϝāĻžāĻ• āĻŦ⧇āϛ⧇ āύāĻŋāϤ⧇ āϏāĻžāĻšāĻžāĻ¯ā§āϝ āĻ•āϰ⧇, āĻļ⧁āϧ⧁ āύāϤ⧁āύāϟāĻŋ āύāϝāĻŧāĨ¤
  • āĻŽāĻžāύ āύāĻŋāϝāĻŧāĻ¨ā§āĻ¤ā§āϰāĻŖ. āĻ­āĻžāϞ⧋ āχāĻžā§āϜāĻŋāύāĻŋāϝāĻŧāĻžāϰāĻŋāĻ‚ āĻĒāĻ›āĻ¨ā§āĻĻ āĻ‰ā§ŽāĻĒāĻžāĻĻāύ⧇ āύāĻŋāĻ°ā§āĻ­āϰāϝ⧋āĻ—ā§āϝāϤāĻžāϰ āϘāϟāύāĻž āĻ•āĻŽāĻŋāϝāĻŧ⧇ āĻĻ⧇āϝāĻŧāĨ¤

āϜāĻ¨ā§āϝ āϏāϤāĻ°ā§āĻ•

  • āĻāĻ•āϟāĻŋ āĻŦ⧇āĻžā§āϚāĻŽāĻžāĻ°ā§āĻ• āĻ…āĻĒā§āϟāĻŋāĻŽāĻžāχāϜ āĻ•āϰāĻž āĻŦ⧃āĻšāĻ¤ā§āϤāϰ āϏāĻŋāĻ¸ā§āĻŸā§‡āĻŽ āĻĻ⧁āĻ°ā§āĻŦāϞāϤāĻž āφāĻĄāĻŧāĻžāϞ āĻ•āϰāϤ⧇ āĻĒāĻžāϰ⧇āĨ¤
  • āĻ…āĻŦāĻ•āĻžāĻ āĻžāĻŽā§‹ āĻāĻŦāĻ‚ āϰāĻ•ā§āώāĻŖāĻžāĻŦ⧇āĻ•ā§āώāϪ⧇āϰ āĻ–āϰāϚ āĻĒā§āϰāĻžāϝāĻŧāχ āĻ…āĻŦāĻŽā§‚āĻ˛ā§āϝāĻžāϝāĻŧāύ āĻ•āϰāĻž āĻšāϝāĻŧāĨ¤
  • āϏāĻŋāĻ¸ā§āĻŸā§‡āĻŽāϗ⧁āϞāĻŋ āφāϰāĻ“ āϜāϟāĻŋāϞ āĻšāĻ“āϝāĻŧāĻžāϰ āϏāĻžāĻĨ⧇ āϏāĻžāĻĨ⧇ āϏ⧁āϰāĻ•ā§āώāĻž āĻāĻŦāĻ‚ āĻĒāĻ°ā§āϝāĻŦ⧇āĻ•ā§āώāĻŖāϝ⧋āĻ—ā§āϝāϤāĻžāϰ āĻĢāĻžāρāĻ• āĻŦāĻžāĻĄāĻŧāϤ⧇ āĻĒāĻžāϰ⧇āĨ¤

āĻĒāϰāĻŦāĻ°ā§āϤ⧀ āĻĒāĻĻāĻ•ā§āώ⧇āĻĒ

  1. āĻŦāĻžāĻ¸ā§āϤāĻŦāĻžāϝāĻŧāύ⧇āϰ āφāϗ⧇ āĻŦāĻŋāϞāĻŽā§āĻŦ, āϗ⧁āĻŖāĻŽāĻžāύ āĻāĻŦāĻ‚ āĻ–āϰāĻšā§‡āϰ āϞāĻ•ā§āĻˇā§āϝ āύāĻŋāĻ°ā§āϧāĻžāϰāĻŖ āĻ•āϰ⧁āύāĨ¤
  2. āĻŦāĻžāĻ¸ā§āϤāĻŦāϏāĻŽā§āĻŽāϤ āϞ⧋āĻĄ āĻāĻŦāĻ‚ āĻĄā§‡āϟāĻž āĻ…āĻŦāĻ¸ā§āĻĨāĻžāϰ āĻ…āϧ⧀āύ⧇ āĻŦ⧇āĻžā§āϚāĻŽāĻžāĻ°ā§āĻ•āĨ¤
  3. āĻ¤ā§āϰ⧁āϟāĻŋ, āĻĒā§āϰāĻŦāĻžāĻš, āĻāĻŦāĻ‚ āĻŦā§āϝāĻŦāĻšāĻžāϰāĻ•āĻžāϰ⧀āϰ āĻĒā§āϰāĻ­āĻžāĻŦ⧇āϰ āϜāĻ¨ā§āϝ āϝāĻ¨ā§āĻ¤ā§āϰ āĻĒāĻ°ā§āϝāĻŦ⧇āĻ•ā§āώāĻŖāĨ¤
  4. āĻ¸ā§āϕ⧇āϞ āĻ•āϰāĻžāϰ āφāϗ⧇ āϰ⧋āϞāĻŦā§āϝāĻžāĻ• āĻāĻŦāĻ‚ āϘāϟāύāĻžāϰ āĻĒā§āϰāϤāĻŋāĻ•ā§āϰāĻŋāϝāĻŧāĻž āĻĒāĻžāĻĨ āĻĒā§āϰāĻ¸ā§āϤ⧁āϤ āĻ•āϰ⧁āύāĨ¤

āĻ…āĻ¨ā§āĻŦ⧇āώāĻŖ āϚāĻžāϞāĻŋāϝāĻŧ⧇ āϝāĻžāύ

āĻĢā§āϰāĻŋ āύāĻŋāωāϜāϞ⧇āϟāĻžāϰ

āĻĒā§āϰāϤāĻŋāĻĻāĻŋāύ⧇āϰ āĻāφāχ āĻŦā§āϰāĻŋāĻĢāĻŋāĻ‚ āĻĒāĻžāύ

āĻĒā§āϰāϤāĻŋ āϏāĻĒā§āϤāĻžāĻšā§‡āϰ āĻ•āĻžāĻœā§‡āϰ āĻĻāĻŋāύ āϏāĻ•āĻžāϞ⧇ āϤāĻŋāύāϟāĻŋ āϝāĻžāϚāĻžāχāĻ•ā§ƒāϤ āĻāφāχ āĻ¸ā§āĻŸā§‹āϰāĻŋ, āϏāĻšāϜ āχāĻ‚āϰ⧇āϜāĻŋāϤ⧇ āϞ⧇āĻ–āĻžāĨ¤ āϚāĻŋāϰāĻ•āĻžāϞ⧇āϰ āϜāĻ¨ā§āϝ āĻĢā§āϰāĻŋ, āϕ⧋āύ⧋ āĻŦāĻŋāĻœā§āĻžāĻžāĻĒāύ āύ⧇āχāĨ¤

āĻĒā§āϰāϤāĻŋ āϏāĻĒā§āϤāĻžāĻšā§‡āϰ āĻ•āĻžāĻœā§‡āϰ āĻĻāĻŋāύ⧇ āĻāĻ•āϟāĻŋ āχāĻŽā§‡āϞāĨ¤ āĻāĻ• āĻ•ā§āϞāĻŋāϕ⧇āχ āφāύāϏāĻžāĻŦāĻ¸ā§āĻ•ā§āϰāĻžāχāĻŦ āĻ•āϰ⧁āύāĨ¤ āφāĻŽāϰāĻž āĻ•āĻ–āύ⧋āχ āφāĻĒāύāĻžāϰ āĻ āĻŋāĻ•āĻžāύāĻž āĻŦāĻŋāĻ•ā§āϰāĻŋ āĻŦāĻž āĻļā§‡ā§ŸāĻžāϰ āĻ•āϰāĻŋ āύāĻžāĨ¤

āύāĻŋāĻœā§‡āϕ⧇ āĻĒāϰ⧀āĻ•ā§āώāĻž āĻ•āϰ⧁āύ

Take the Data-Constrained Scaling Laws quiz

āĻĒā§āϰāϤāĻŋāϟāĻŋ āωāĻ¤ā§āϤāϰ⧇āϰ āϤāĻžā§ŽāĻ•ā§āώāĻŖāĻŋāĻ• āĻĢāĻŋāĻĄāĻŦā§āϝāĻžāĻ•, āĻāĻŦāĻ‚ āϕ⧋āĻ°ā§āϏ āĻĒāĻžāϏ āĻ•āϰāĻžāϰ āĻĒāϰ āϝāĻžāϚāĻžāχāϝ⧋āĻ—ā§āϝ āφāχāĻĄāĻŋ āϏāĻš āĻāĻ•āϟāĻŋ āĻļā§‡ā§ŸāĻžāϰāϝ⧋āĻ—ā§āϝ āϏāĻžāĻ°ā§āϟāĻŋāĻĢāĻŋāϕ⧇āϟāĨ¤

āϕ⧁āχāϜ āĻļ⧁āϰ⧁ āĻ•āϰ⧁āύ

āĻĢā§āϰāĻŋ āĻāφāχ āĻļāĻŋāĻ•ā§āώāĻžāϕ⧇ āϏāĻŽāĻ°ā§āĻĨāύ āĻ•āϰ⧁āύāĨ¤ AI Understanding āĻāĻ•āϟāĻŋ 501(c)(3) āĻ…āϞāĻžāĻ­āϜāύāĻ• āĻĒā§āϰāϤāĻŋāĻˇā§āĻ āĻžāύ — āϕ⧋āύ⧋ āĻŦāĻŋāĻœā§āĻžāĻžāĻĒāύ āύ⧇āχ, āϕ⧋āύ⧋ āĻĒ⧇āĻ“ā§ŸāĻžāϞ āύ⧇āχ, āĻ•āĻ–āύ⧋āχ āύāĻžāĨ¤ āĻāĻ•āϟāĻŋ āĻĻāĻžāύ āĻ•āϰ⧁āύ

āĻĒā§āϰāĻžāϝāĻŧāĻļāχ āϜāĻŋāĻœā§āĻžāĻžāϏāĻŋāϤ āĻĒā§āϰāĻļā§āύāĻžāĻŦāϞ⧀

What is Data-Constrained Scaling Laws?

Data-constrained scaling studies how to allocate training resources when the supply of usable unique data is limited. Repeating a corpus increases processed tokens without creating the same amount of new information. A useful plan distinguishes unique tokens, repeated exposures, model size, and evidence from held-out evaluation.

A corpus contains ten billion unique tokens and is processed for three complete epochs. Which accounting is correct?

Repeated passes add training exposures without increasing the unique corpus size: 10 billion × 3 = 30 billion.

Why can two runs with the same processed-token count learn differently?

The guide distinguishes unique information from repeated exposures and recommends evaluating the resulting models.

How should the cited study’s finding about roughly four epochs be interpreted?

The guide scopes the reported small loss differences to the study’s configurations and calls for local experiments.

What does a fitted effective-data quantity represent in this setting?

Effective data summarizes estimated learning value under the fitted model; it is not the physical corpus count.

Training loss keeps falling while held-out loss rises after more repetition. Which response follows the guide?

Worsening held-out behavior is a reason to investigate and compare checkpoints rather than blindly add compute.

āĻļāĻŋāĻ–āϤ⧇ āĻĨāĻžāϕ⧁āύ

āϏāĻŽā§āĻĒāĻ°ā§āĻ•āĻŋāϤ āĻ—āĻžāχāĻĄ

āĻāχ āĻŦāĻŋāώāϝāĻŧ⧇āϰ āϜāĻ¨ā§āϝ āĻŦāĻžāĻ›āĻžāχ āĻ•āϰāĻž āφāϰāĻ“ āĻ—āĻžāχāĻĄ

āĻĒāϰāĻŦāĻ°ā§āϤ⧀ āφāĻĒāĻĒāϰāĻŦāĻ°ā§āϤ⧀ āĻ—āĻžāχāĻĄ

āϚāĻŋāύāϚāĻŋāϞāĻž āĻ¸ā§āϕ⧇āϞāĻŋāĻ‚ āφāχāύ

āĻ­āĻžāώāĻž āĻāφāχ

āĻāĻ•āϟāĻŋ āĻŦāĻŋāύāĻžāĻŽā§‚āĻ˛ā§āϝ⧇ AI āĻļāĻ‚āϏāĻžāĻĒāĻ¤ā§āϰ āĻĒāĻžāύāĻāĻ•āϟāĻŋ āϏāĻ‚āĻ•ā§āώāĻŋāĻĒā§āϤ āĻŦāĻŋāύāĻžāĻŽā§‚āĻ˛ā§āϝ⧇ āϕ⧋āĻ°ā§āϏ āĻāĻŦāĻ‚ āĻĒāϰ⧀āĻ•ā§āώāĻž. āϕ⧋āύ⧋ āĻ–āϰāϚ āύ⧇āχ, āϕ⧋āύ⧋ āĻ•āĻžāĻ°ā§āĻĄ āύ⧇āχāĨ¤ āĻĒā§āϰāĻžāϝāĻŧ 2.5 āϘāĻ¨ā§āϟāĻžāĨ¤āĻŦāĻŋāύāĻžāĻŽā§‚āĻ˛ā§āϝ⧇ āĻļ⧁āϰ⧁ āĻ•āϰ⧁āύ →