INDEX

Explanations

terms related to someone being angry or outraged

oai_token-act-pair · gpt-3.5-turbo

instances of the prefix "ir," indicating negation or reversal

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.8.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

 guiActiveUnfocused

-0.81

hetti

-0.66

 Franks

-0.65

coat

-0.64

 Order

-0.63

sheet

-0.63

lished

-0.62

erest

-0.61

ezvous

-0.60

 Ragnarok

-0.60

POSITIVE LOGITS

idium

1.03

cles

1.01

relevant

0.99

ked

0.97

respective

0.96

religious

0.93

rational

0.88

ruption

0.87

respons

0.87

ansas

0.81

Activations Density 0.023%

terms related to someone being angry or outraged

instances of the prefix "ir," indicating negation or reversal

No Comments

No Known Activations