INDEX

Explanations

actions related to watching something either actively or passively

oai_token-act-pair · gpt-3.5-turbo

phrases related to watching or viewing activities

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

esan

-0.89

uti

-0.79

ãĥ´

-0.79

hesion

-0.75

ItemTracker

-0.73

pmwiki

-0.71

eno

-0.70

enos

-0.69

interstitial

-0.69

quart

-0.69

POSITIVE LOGITS

 television

1.07

TV

1.05

tv

1.05

 helpless

1.05

dogs

1.03

 unfold

1.02

 videos

1.01

 documentaries

0.99

 closely

0.98

 movies

0.95

Activations Density 0.080%

actions related to watching something either actively or passively

phrases related to watching or viewing activities

No Comments

No Known Activations