INDEX

Explanations

hyphenated words that signal opting in or out of something

oai_token-act-pair · gpt-3.5-turbo

occurrences of phrases indicating the option to opt out

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

 Lawrence

-0.76

Lod

-0.76

 pulp

-0.76

 Norris

-0.71

 Revis

-0.71

Moj

-0.70

 clenched

-0.69

 Galile

-0.67

 Arabian

-0.67

 Guinness

-0.65

POSITIVE LOGITS

direction

1.12

cert

1.09

cycle

1.03

eligible

1.01

friendly

0.98

equipped

0.98

contract

0.98

based

0.98

items

0.98

tun

0.97

Activations Density 0.076%

hyphenated words that signal opting in or out of something

occurrences of phrases indicating the option to opt out

No Comments

No Known Activations