INDEX

Explanations

various occurrences of the word "head" at differing activation levels

oai_token-act-pair · gpt-3.5-turbo

references to leadership positions or titles

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

PsyNetMessage

-0.70

Avg

-0.69

 resil

-0.67

 mell

-0.67

 Strat

-0.61

 estab

-0.61

Suc

-0.60

ĸļ

-0.60

Constructed

-0.60

Mub

-0.59

POSITIVE LOGITS

canon

1.10

quarter

1.07

scar

0.98

heading

0.98

gear

0.97

phones

0.96

butt

0.96

jack

0.93

lining

0.92

liner

0.91

Activations Density 0.025%

various occurrences of the word "head" at differing activation levels

references to leadership positions or titles

No Comments

No Known Activations