AlphaCheck

See which layers of your fine-tune are over- or under-trained

Paste a public Hugging Face model. AlphaCheck fits a power law to the eigenvalue spectrum of every weight matrix and flags each layer underfit, healthy or overfit. No eval data, no GPU, nothing to install.

Public, ungated text transformer up to 13B parameters, in safetensors format.
Compare a fine-tune with its base model, layer by layer, with a subscription ($9/month).
3 free reports left, no sign-up.

Try:

A real report: Qwen/Qwen2.5-0.5B-Instruct, all 168 weight matrices from layer 0 to 23 Mean α 5.90; 96 healthy, 67 underfit, 5 overfit. Computed November 22, 2026 at commit 7ae5576. Download its CSV.
overfit (α < 2)healthy (2 to 6)underfit (α > 6)▲ above the chart top

How to read α

α < 2, overfitThe spectrum is very heavy-tailed: the layer may have memorised its training data.
2 ≤ α ≤ 6, healthyThe range well-trained layers fall in.
α > 6, underfitClose to random: the layer has learned little structure.

Ranges are WeightWatcher's published thresholds. The health score is the share of matrices in the healthy range.

Method

For each weight matrix W in the transformer blocks (attention and MLP projections), we compute the eigenvalues of WᵀW and fit p(λ) ∝ λ−α to the tail, choosing the tail start by the smallest Kolmogorov-Smirnov distance.

This is Heavy-Tailed Self-Regularization, from Martin, Peng & Mahoney, Nature Communications (2021), with the power-law fit of Clauset, Shalizi & Newman (2009). The open-source WeightWatcher does the same analysis on your own machine; AlphaCheck runs it for you on our server, straight from the Hub.

Weights stream one matrix at a time from Hugging Face. A 0.5B model takes about half a minute; a 7B model takes around half an hour on our current server, and the report fills in as it goes.

What it can analyze

  • Public, ungated text models on Hugging Face, up to 13B parameters
  • Weights in safetensors format (bf16, fp16 or fp32)
  • Full models, including merged fine-tunes. LoRA-only adapters, vision and multimodal models, and non-transformer architectures are not supported.

Subscribers can add a base model to see the α change of every matrix, which shows where fine-tuning moved the weights. $9/month, cancel anytime. The first 3 reports are free.