INDEX

Explanations

positively or negatively charged adjectives or phrases indicating approval or disapproval

oai_token-act-pair · gpt-3.5-turbo

expressions related to the concepts of good and better, as well as phrases indicating a moral evaluation

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

ned

-0.57

gall

-0.56

 Glac

-0.56

 Notting

-0.55

 Clancy

-0.53

pat

-0.53

egu

-0.52

 Guam

-0.52

 Rolls

-0.52

 Tornado

-0.51

POSITIVE LOGITS

 sake

1.62

 purposes

1.35

 reasons

1.21

ummies

1.20

 reason

0.94

 Reasons

0.90

aughs

0.87

 purpose

0.87

 ages

0.84

instance

0.80

Activations Density 0.187%

positively or negatively charged adjectives or phrases indicating approval or disapproval

expressions related to the concepts of good and better, as well as phrases indicating a moral evaluation

No Comments

No Known Activations