INDEX

Explanations

words or suffixes related to medical conditions or professions

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_post_32k-oai

Configuration

jbloom/GPT2-Small-OAI-v5-32k-resid-post-SAEs/v5_32k_layer_8.pt

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

32,768

Data Type

torch.float32

Hook Name

blocks.8.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

ers

-0.71

orem

-0.65

eny

-0.64

-0.63

encers

-0.63

erest

-0.62

ership

-0.61

ured

-0.57

 Orwell

-0.57

orses

-0.57

POSITIVE LOGITS

clips

1.00

idge

0.98

jee

0.98

bilt

0.95

lein

0.95

geist

0.91

rors

0.88

ror

0.87

chief

0.86

lain

0.84

Activations Density 0.451%

words or suffixes related to medical conditions or professions

No Comments

No Known Activations