INDEX

Explanations

Twitter usernames

oai_token-act-pair · gpt-3.5-turbo

underscore characters or special formatting

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

atform

-0.74

 Levine

-0.73

 Holding

-0.69

 Frazier

-0.69

 Pearce

-0.69

FML

-0.68

 Manson

-0.68

Ago

-0.68

quished

-0.68

 Stef

-0.67

POSITIVE LOGITS

default

1.06

dict

1.02

chance

1.00

blank

0.99

EStreamFrame

0.97

tro

0.95

events

0.94

gradient

0.94

token

0.92

delay

0.92

Activations Density 0.024%

Twitter usernames

underscore characters or special formatting

No Comments

No Known Activations