01 READOUT
Lens
What is represented inside?
Decode residual states, attention-head outputs, and projected visual embeddings into vocabulary predictions.
Lens method guidesA TOOLKIT FOR MODEL INTERPRETABILITY
vlmprobe
Beyond the answer.
Inside the model.
Explore internal predictions, attention, causal dependence, and sparse features in language, vision, and multimodal models.
Lens readouts · Attention explanations · Causal interventions · Sparse features
A LOOK INSIDEActual model measurements
GQA / 07302654
“Is the shirt black or yellow?”
Generated answerYellow
What happens when attention routes are blocked?
Image to questionImage to answerQuestion to answer
One model binding.
Four complementary views.
04method families
08VLM adapter families
09live walkthroughs
01 / OVERVIEWExploring model internals beyond the answer
An answer is only
the beginning.
vlmprobe connects observations and interventions through one interface bound to a model you already own.
Read vocabulary predictions from internal states. Compare attention patterns on the same input. Intervene on activations or attention routes and measure what changes. Train or load sparse dictionaries to inspect individual features. Explicit adapters handle model-specific hook locations, visual-token coordinates, and readouts.
Each result keeps tensors and experiment metadata together, so the same computation can be inspected, visualized, saved, and replayed.
02 / METHODSRead. Attribute. Intervene. Decompose.
Different questions.
One probing interface.
From internal predictions to controlled changes,
choose the view your research question needs.
02 EXPLANATION
Attention
Where does the model look?
Compare attention flow and target-conditioned spatial attributions on the same image and question.
Attention method guides03 INTERVENTION
Causal
Which computations matter?
Patch activations, isolate paths, block attention routes, or steer residual states and measure the effect.
Causal method guides04 SPARSE FEATURES
SAE
Which sparse features activate?
Train or load dictionaries, locate activating tokens or image regions, and measure feature edits with reconstruction controls.
SAE method guidesPretrained model cardsMethod availability depends on the model adapter. Fitted lenses and SAE dictionaries require matching model, layer, and input conventions.
03 / FRAMEWORKModel-bound by design
Bind once. Probe together.
A shared execution contract connects native inputs,
explicit adapters, and structured results.
01
Explicit adapters
Architecture-specific hooks and expanded visual-token coordinates.
02
Capability checks
Unsupported operations are distinguished from measured zero effects.
03
Structured results
Tensors, metadata, and shared visualizations travel together.
04 / EXPLORE AN EXPERIMENTArchived measurements · no model loading
Block a route.
Watch the answer change.
Replay a real Attention Knockout scan.
Different routes have different effects on the same input.
Answer-token probability
MEASURED REPLAYAfter route blockingUnmodified baseline
Loading the archived measurement…
Experimental settings & source
Joint windows of up to 9 consecutive layers, clipped at the boundaries; all heads are blocked at masked pre-softmax attention logits. Layer indices start at zero. LLaVA-1.5-7B uses bf16 model execution and float32 probability readout.
This browser replays stored numbers; it does not run a new intervention. A single GQA example does not establish dataset-level performance.
View original measurementNine notebooks. Three perspectives.
Browse all demonstrations
LENS
Text lens comparison
GPT-2 · inputs, calibration & readouts
LENSVLM lens comparison
SmolVLM · image inputs & lens calibration
ATTENTIONViT attention comparison
ViT-B/16 · input-to-map walkthrough
ATTENTIONVLM attention comparison
SmolVLM · complete-answer attribution
CAUSALActivation & attribution patching
GPT-2 · measured & estimated patch effects
CAUSALAttention Knockout
SmolVLM · live route blocking
CAUSALPath patching
GPT-2 · controlled receiver replay
CAUSALResidual steering / VSV
SmolVLM · live captions & strength responses
CAUSALEAP / EAP-IG
GPT-2 · edge scores & circuit recovery
Notebooks prepare raw inputs, run the library and draw fresh results. Calibrated lenses include training steps; costly optional sections are labeled explicitly.
05 / MODEL SUPPORTExplicit architecture contracts
Across vision and language.
Eight VLM families, with language and vision controls.
Check capabilities for your specific model and method.
COMPLEMENTARY CONTROLSGPT-2LlamaViT-B/16
Support is method-specific.
Adapter support and pretrained validation are different scopes. Five pretrained VLMs were exercised in the technical report; some families have architecture-level validation. Qwen3.5 has restrictions across its hybrid attention stack. Fitted decoders are not supplied for every model.
Read the capability matrix06 / GET STARTEDPython 3.10+ · PyTorch
Your model.
A clearer view.
Install from PyPI, bind a loaded model, and inspect the methods available to it.
Read the API guideModel weights and complete datasets are downloaded separately.
INSTALL & PROBE
# Install from PyPI
python -m pip install "vlmprobe[transformers,visualization]"
# In Python, with your loaded model and native inputs
from vlmprobe import Prober
probe = Prober(model, processor)
print(probe.describe())
result = probe.lens.logit(
layers=[0, 13, 27],
tokens="last_prompt",
).run(inputs)
result.save("outputs/logit_lens")07 / RESOURCESContinue exploring
REPORT
Technical report
Shared-case comparisons across five pretrained VLMs, with language and vision controls.
DOCUMENTATIONGuides & API
Model loading, method definitions, calibration workflows, and visualization tools.
PROVENANCEPapers & datasets
Original methods, author implementations, and sources used for the demonstrations.