INDEX

Explanations

words or phrases related to communication actions, such as calling, emailing, and messaging

oai_token-act-pair · gpt-3.5-turbo

instances of the word "and."

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 9-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.9.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.9.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

mite

-0.71

utenberg

-0.69

LOCK

-0.63

adish

-0.63

aepernick

-0.63

gur

-0.61

 disproportion

-0.61

 hiber

-0.60

 disproportionate

-0.60

 attRot

-0.60

POSITIVE LOGITS

 asked

1.16

 inquired

1.15

 asks

1.03

 begged

1.02

 congratulated

0.98

 thanked

0.96

ask

0.95

 apologized

0.94

 gave

0.91

 told

0.91

Activations Density 0.221%

words or phrases related to communication actions, such as calling, emailing, and messaging

instances of the word "and."

No Comments

No Known Activations