INDEX

Explanations

links to images, specifically Twitter image links

oai_token-act-pair · gpt-3.5-turbo

punctuation marks, specifically periods

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

 audits

-0.69

 involuntary

-0.66

 volunt

-0.64

 confidentiality

-0.62

 indemn

-0.62

 tsun

-0.61

 conformity

-0.60

 equilibrium

-0.60

 interchange

-0.60

 disadvant

-0.59

POSITIVE LOGITS

twitter

1.66

facebook

1.13

imgur

1.08

google

1.05

twitch

1.02

redd

0.94

youtube

0.93

reddit

0.90

blogspot

0.90

github

0.87

Activations Density 0.026%

links to images, specifically Twitter image links

punctuation marks, specifically periods

No Comments

No Known Activations