INDEX

Explanations

instances of laughter or comedic responses

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 10-res_post_32k-oai

Configuration

jbloom/GPT2-Small-OAI-v5-32k-resid-post-SAEs/v5_32k_layer_10.pt

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

32,768

Data Type

torch.float32

Hook Name

blocks.10.hook_resid_post

Hook Layer

Architecture

standard

Context Size

Dataset

Skylion007/openwebtext

Activation Function

topk

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

roman

-0.73

glers

-0.72

indal

-0.65

CGI

-0.65

 mathemat

-0.65

 quotas

-0.63

ailability

-0.63

geries

-0.63

knife

-0.63

document

-0.62

POSITIVE LOGITS

Laughs

1.00

 nervously

0.92

laughs

0.92

 maniac

0.80

 Slightly

0.78

laugh

0.75

ause

0.72

 hung

0.72

hum

0.71

 softly

0.70

Activations Density 0.259%

instances of laughter or comedic responses

No Comments

No Known Activations