INDEX

Explanations

contractions combined with the negation "not"

oai_token-act-pair · gpt-3.5-turbo

negations or phrases indicating disagreement

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

 fixme

-0.68

ership

-0.66

IER

-0.65

 Evaluation

-0.64

velt

-0.61

ilage

-0.61

inav

-0.61

ilege

-0.60

 decency

-0.59

cano

-0.59

POSITIVE LOGITS

 alone

1.29

shy

1.17

 amused

1.07

 afraid

1.02

 immune

1.01

 necessarily

0.96

 exactly

0.93

 ashamed

0.92

 Alone

0.92

 kidding

0.91

Activations Density 0.103%

contractions combined with the negation "not"

negations or phrases indicating disagreement

No Comments

No Known Activations