INDEX

Explanations

names of organizations or positions followed by some context or interaction

oai_token-act-pair · gpt-3.5-turbo

references to law enforcement or legal contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.11.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 sidx

-0.76

Jews

-0.75

 embryos

-0.75

rooms

-0.72

actly

-0.70

 rigged

-0.67

eches

-0.67

 prizes

-0.66

successfully

-0.66

ouls

-0.65

POSITIVE LOGITS

 spokesman

1.36

 spokeswoman

1.34

 spokesperson

1.33

 Director

1.13

 Chief

1.11

 Deputy

1.10

 Superintendent

1.08

 Assistant

1.05

 deputy

1.05

 Principal

1.03

Activations Density 0.208%

names of organizations or positions followed by some context or interaction

references to law enforcement or legal contexts

No Comments

No Known Activations