INDEX

Explanations

words related to legal matters and political figures in news contexts

oai_token-act-pair · gpt-3.5-turbo

terms related to elements and their occurrences

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

 America

-0.68

 Craigslist

-0.63

 Omaha

-0.63

 kills

-0.62

 dirty

-0.62

 offensive

-0.60

 kill

-0.60

 papers

-0.60

bro

-0.60

 Grind

-0.59

POSITIVE LOGITS

lement

5.03

lements

2.58

lem

1.21

ling

1.09

ment

1.06

lers

1.04

LE

1.03

lez

1.01

LES

1.00

ement

0.99

Activations Density 0.014%

words related to legal matters and political figures in news contexts

terms related to elements and their occurrences

No Comments

No Known Activations