NeuroNomixer
  • Home
  • Blog
  • Visual Guides
  • Authors
  • Contact
Sign InSign Up
HomeBlogAuthorsContactPrivacy Policy

© 2026 NeuroNomixer — Built with Next.js & Tailwind CSS

Visual Guides/Scaling Laws
LLMs

Scaling Laws

Loss falls predictably with parameters, data, and compute, which is why frontier labs can plan a model before training it. Here you drive the actual fitted Chinchilla law: pick a FLOPs budget, split it between model size and tokens, and read the predicted loss live.

Change the compute budget
Move the N vs D allocation
Inspect a historical model

Sign in to save progress

The law, with its published constants

Kaplan et al. 2020 (arXiv:2001.08361) showed language model loss follows smooth power laws in model size, data, and compute. Hoffmann et al. 2022 (arXiv:2203.15556, the Chinchilla paper) fitted a two-term parametric form to hundreds of training runs. Every number in this guide is computed from that fit:

L(N, D) = E + A / N^alpha + B / D^beta

E = 1.69      A = 406.4     B = 410.7
alpha = 0.34          beta = 0.28

constants: Hoffmann et al. 2022, arXiv:2203.15556 (Approach 3 fit)
compute:   C = 6 * N * D FLOPs (Kaplan et al. 2020 approximation)

E: the floor

The irreducible loss. Even an infinite model on infinite data cannot beat the intrinsic entropy of natural text. No amount of scale buys past it.

A / N^alpha: model term

The penalty for a finite parameter count. It shrinks as a power law in N, so every extra decade of parameters buys a predictable, and shrinking, slice of loss.

B / D^beta: data term

The penalty for finite training data. Because both terms decay as power laws, a fixed compute budget has a best split between the two: the compute-optimal allocation.

Spend a budget: the iso-FLOP valley

Fix a compute budget C and every choice of N drags D = C / 6N with it: a bigger model must train on fewer tokens. The curve below is the predicted loss along that constraint. Its valley is the compute-optimal point. Slide the budget and the whole valley shifts; slide the allocation and watch how much loss a lopsided split costs you.

1.0e23 FLOPs

Log scale, 1e19 to 1e26 FLOPs. The valley of the curve is the compute-optimal allocation for this budget.

+0.00 dec

0.00 is the compute-optimal N. Negative: smaller model, more tokens. Positive: bigger model, fewer tokens. C stays fixed.

1.881.992.102.202.311B10B100BParameters N (log scale)Predicted lossoptimalyoudashed: C / 10 and C x 10

Parameters N

14.6B

Tokens D = C/6N

1.14T

Tokens per param

78.2

Predicted loss

2.005

Optimal loss

2.005

Loss overhead

+0.000 (0.0%)

Frontier allocation at this budget: 14.6B parameters on 1.14T tokens (78 tokens per parameter). All values are computed live from L(N, D) with the Hoffmann et al. 2022 constants.

The frontier, and where real models landed

Sweeping the budget traces the compute-optimal frontier through the (N, D) plane. The purple dots are real published models, plotted at their published parameter and token counts. GPT-3-era models sit far to the parameter-heavy side of the frontier; Chinchilla was the correction. Llama-generation models are famously trained far past the 20-tokens-per-parameter rule of thumb because inference cost rewards small models, yet against the more data-heavy Approach 3 frontier drawn here even they land on the parameter-heavy side, a tension the fine print below unpacks.

1B10B100B1T10T100T100M1B10B100B1T10TParameters N (log scale)Training tokens D (log scale)1e20 FLOPs1e22 FLOPs1e24 FLOPs1e26 FLOPscompute-optimal frontierGPT-3GopherMT-NLGChinchillaPaLMLLaMA 65BLlama 2 70BLlama 3.1 405Byou

Inspect a model

Select a model above to see its published parameter and token counts, its computed budget, and how far the Chinchilla fit places it from the compute-optimal frontier.

Read the fine print

  • Constants are dataset-specific. The fit was calibrated on DeepMind's MassiveText with their tokenizer. Predicted losses for other models show the shape of the tradeoff, not those models' actual reported losses.
  • The published fit is contested. These Approach 3 constants imply a frontier of roughly 30 to 150 tokens per parameter across this guide's budget range (computed live above), which is more data-heavy than the famous 20-tokens-per-parameter headline from the same paper's Approaches 1 and 2. Besiroglu et al. 2024 (arXiv:2404.10102) document the inconsistency and refit the constants.
  • C = 6ND is an approximation. It counts dense forward and backward FLOPs per token and ignores attention overhead, embedding parameters, and architecture details.
  • Compute-optimal is not deployment-optimal. The law prices training only. A model served billions of times is often trained far past the frontier on purpose: extra training FLOPs buy a smaller model with cheaper inference forever.
← All GuidesNext Guide →