p.enthalabs

nanointerpret

How do you want to poke at the LLM?

I want to force activations Force a feature up and see how it changes the generated text I want to explore the activations Browse the concepts the SAE has learned and see what makes each one fire I want to do both Do both intervention and feature exploration

nanointerpret

[](https://github.com/Belluxx/nanointerpret)

Model Loading...

Change mode

Intervention playground

Force the activation of a feature to see how it influences the text generated by the model

? Sampling

How to use the playground

1. Enter a prompt like I was driving when 2. Choose the feature that you want to forcefully activate. Open the feature dropdown and scroll or search by ID or title. 3. Choose how strongly to activate the feature:

- **50%** mildly activated

- **100%** maximum observed activation

- **200%** twice the maximum (may break the model)

4. Press **Generate comparison**

Sampling settings

Maximum tokens

Temperature

Top P

Top K

Repetition penalty

Examples

Prompt

Feature

Select a feature

No features found

Target activation (% of feature max)

50%100%150%200%

Generate comparison

Unmodified

Intervened

Filters

Category

Semantic

All categories

Token-specific

Lexical

Semantic

Activation count

Clear

Sort

Activation count

Feature ID

Activation count

Title

Loading activations...High quality Activations

Page