INDEX

Explanations

phrases and structures indicating purpose or intent related to actions

oai_token-act-pair · gpt-4o-mini Triggered by @bot

purpose or function descriptions

np_acts-logits-general · gemini-2.5-flash-lite

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k

Configuration

google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

131,072

Data Type

float32

Hook Name

blocks.20.hook_resid_post

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 alguno

-0.52

 kysy

-0.45

 franceses

-0.44

 algum

-0.43

 colgantes

-0.42

ถาม

-0.40

 inesperado

-0.40

 vicenda

-0.40

Litteratur

-0.40

 bertanya

-0.40

POSITIVE LOGITS

 Designed

0.57

Designed

0.57

 designed

0.56

 Prun

0.53

 pigmented

0.51

 optimally

0.49

 maximize

0.49

 destined

0.48

 intrinsically

0.48

 maximized

0.47

Activations Density 0.023%

phrases and structures indicating purpose or intent related to actions

purpose or function descriptions

No Comments

No Known Activations