INDEX

Explanations

words related to specific names, potentially from social media posts or news articles

oai_token-act-pair · gpt-3.5-turbo

words containing the suffix "ette."

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 7-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.7.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.7.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

committee

-0.87

ifted

-0.77

arcity

-0.77

ansom

-0.76

rush

-0.76

ific

-0.75

reddits

-0.73

ardless

-0.73

 glim

-0.72

iant

-0.72

POSITIVE LOGITS

lla

0.86

ette

0.81

vre

0.80

 Scotia

0.77

 Hebdo

0.75

ptin

0.74

Mania

0.74

brate

0.74

phrine

0.74

brates

0.72

Activations Density 0.031%

words related to specific names, potentially from social media posts or news articles

words containing the suffix "ette."

No Comments

No Known Activations