INDEX

Explanations

occurrences of the word "whe" followed by numbers

oai_token-act-pair · gpt-3.5-turbo

references to specific names or terms related to people or brands

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

istor

-0.82

nick

-0.81

nikov

-0.74

 drown

-0.69

 Saban

-0.68

 COVER

-0.66

 underwater

-0.66

 paras

-0.66

 underworld

-0.66

 frag

-0.65

POSITIVE LOGITS

Whe

3.94

Whe

3.29

whe

3.03

whe

2.77

 Wheat

1.47

WH

1.19

Ye

1.06

Soy

1.02

 Wheels

1.02

Pow

0.99

Activations Density 0.038%

occurrences of the word "whe" followed by numbers

references to specific names or terms related to people or brands

No Comments

No Known Activations