CONTEXT JAMMING

Field notes from inside the context window.

Weight-Space Geometry

The Sixteen-Joint Machine

The useful content of a trained network doesn’t live in all its weights — it lives in about sixteen directions per layer, joints that every model, regardless of task or initialization, converges onto. The open question is no longer whether that subspace exists, but whether we can pretrain natively inside it instead of rediscovering it from scratch on every run.

01 · The finding in 60 seconds

An action figure has about sixteen points of articulation

A collectible action figure isn’t posable because of the plastic — it’s posable because of a small number of engineered joints molded into each limb: shoulders, elbows, hips, a handful of others, usually around sixteen. Every pose the figure can strike is just a setting on those same joints.

The paper’s discovery is that trained neural networks look the same way, except nobody designed the joints on purpose — they emerge from training itself. Across 1,100+ models — ViTs, LLaMA3, GPT-2, Mistral LoRAs, and ResNet-50s trained from scratch on entirely disjoint data — the majority of each layer’s useful weight variance concentrates into roughly sixteen shared directions. Different tasks, different random initializations, even different training datasets: same joints, different pose. Most strikingly, 500 randomly initialized ViTs, never trained at all, already sit on the same low-rank subspace.

How they found it

Truncated, zero-centered Higher-Order SVD (HOSVD) across each population of models, framed theoretically as a Hilbert-space second-moment operator. A two-level convergence theorem bounds the subspace error as it shrinks at O(1/√T) in the number of tasks analyzed, plus a per-task estimation term; how tightly that bound holds is governed by the eigengap γk between the k-th and (k+1)-th eigenvalues (Davis–Kahan). All of it ran on a single NVIDIA A5000 — extraction carries no training costs whatsoever.

02 · Where the joints are

Explained variance drops off a cliff around 16

Drag k to change how many of the top directions you keep. Almost everything worth having is captured early; almost everything past ~16 is what the paper calls the secondary subspace — directions that, when used on their own, perform drastically worse.

11650100
Principal component index100 total
Retained

81.9%

of variance in the first 16 joints

Secondary subspace

18.1%

discarded — the leftover directions the paper shows perform drastically worse

Illustrative scree profile synthesized to match the paper’s reported shape: a sharp drop across the first ~16 components, then a long near-zero tail. Drag the slider to change k and watch how little the remaining components add.

03 · Pose the figure

Sixteen sliders, one shared skeleton

Each slider, c1 through c16, is a coefficient on one of the sixteen shared joints driving a live rank-16 reconstruction of an abstract weight-pattern grid. The presets show what the paper means by “different tasks are different poses”: same joints, entirely different fingerprint.

c10.80
c20.30
c3-0.20
c40.10
c50.50
c6-0.40
c70.20
c80.10
c9-0.10
c100.30
c110.10
c12-0.20
c130.05
c140.10
c15-0.05
c160.10
Sixteen sliders, one fixed set of joints. Each preset is just a different coefficient fingerprint on the same rank-16 basis — the way the paper describes different finetuned tasks as different poses on a shared, architecture-specific subspace.

04 · The evidence

What holds up when you check the numbers

OOD projection & reconstruction

Unseen models, reconstructed inside the 16-dimensional subspace, checked against their full-parameter originals.

94.1% vs 94.4%

ViT · IID accuracy (16-dim vs full)

87.8% vs 91.3%

ViT · OOD accuracy (16-dim vs full)

5 unseen ViTs reconstructed in the subspace.

ResNet-50, five datasets, full → Universal R50

Average across 5 datasetsFull model 90.89Universal subspace 87.58
ImageNetFull model 80.86Universal subspace 77.89
EuroSATFull model 98.96Universal subspace 98.83
CIFAR-10Full model 97.35Universal subspace 95.89
CIFAR-100Full model 83.82Universal subspace 81.49
Oxford-IIIT PetsFull model 93.48Universal subspace 83.81

Coefficient-only training

Training just the ~16 coefficients per layer, with the shared joints frozen.

~10K

Trainable params, Universal ViT

vs 86M for the full model

84.01

GLUE avg, RoBERTa-base universal order-2

At 2× speedup vs LoRA's 83.67

19×

Mistral LoRA subspace memory efficiency

Universal ViT (~10K params) vs full ViT (86M params)

CIFAR-100Full 92.8Universal 90.1
Food101Full 90.7Universal 89.1
Flowers102Full 98.82Universal 90.1
CIFAR-10Full 99.0Universal 96.7
Oxford-IIIT PetsFull 91.2Universal 89.4

Model merging

Eight ViT tasks, merged into one model, scores normalized.

83.5

Ours, average across 8 tasks

68.0

Best baseline (KnOTS-TIES), average

88.1

Cars

82.3

DTD

65.9

EuroSAT

61.3

GTSRB

88.3

MNIST

98.1

RESISC45

98.5

SUN397

85.1

SVHN

Compression

Storing a population of models as one shared basis plus small per-model coefficients.

150 GB

500 ViTs, compressed

>100× cut

1.6 TB

LLaMA collection, compressed

>100× cut

19×

Mistral LoRA memory efficiency

19.83 vs 19.73

SDXL universal subspace CLIP score

Top 16 components, 30 denoising steps, vs individual style LoRAs

Ablation — diminishing returns

Mistral OOD Rouge-L, as the universal subspace is estimated from more source models.

73.7

Normal (full)

55.8

From 50 models

66.1

From 150

71.9

From 250

72.3

From 450

05 · Why it matters

Three implications, in order of how settled they are

Settled

Compression and reuse

Training ~10K coefficients instead of 86M parameters, storing 500 models in 150 GB instead of the original size, cutting Mistral LoRA memory by 19× — this is demonstrated, working today.

Open

The unbuilt leapfrog

Every universal subspace so far was extracted after training. The authors flag native subspace pretraining — building a model inside the 16 joints from the start — as future work, not something anyone has done.

Concerning

The monoculture risk

If every model collapses onto the same subspace, they may inherit the same biases and failure modes. The authors raise this themselves: is shared-basis convergence a diversity bottleneck?

06 · Honest limitations

Where to stay skeptical

  • Shared-checkpoint critique. The ~500 ViTs pulled from HuggingFace were unfiltered, with no access to training data or provenance — they may share base checkpoints, which would make convergence look more universal than it is.
  • Thin from-scratch evidence. The strongest from-scratch claim rests on 5 ResNet-50s trained on disjoint datasets and 500 randomly initialized (untrained) ViTs — a small sample for a universal claim.
  • Interpretability is unproven. The paper itself describes interpreting what the 16 directions actually represent as “extremely cumbersome.” Nobody has confirmed what each joint does.
  • Nobody trains inside the subspace yet. Model-independent, data-derived universal bases, and native pretraining inside them, remain future work — not a demonstrated result.

07 · Field guide

A compact glossary

Weight subspace

A low-dimensional set of directions in a layer's weight space that captures most of the variance across many trained models.

HOSVD (truncated, zero-centered)

Higher-Order Singular Value Decomposition — the tensor generalization of SVD the paper uses to find the shared low-rank directions across a population of models.

Coefficients (a pose)

The small set of numbers (here, ~16 per layer) that place one specific model at one specific point within the shared subspace.

Secondary subspace

The leftover, discarded directions beyond the top ~16 — shown to perform drastically worse, which is part of the evidence the chosen 16 actually matter.

Eigengap (γk)

The gap between the k-th and (k+1)-th eigenvalues of the second-moment operator; the paper's stability guarantee gets weaker as this gap shrinks.