INDEX

Explanations

phrases indicating instructions or recommendations

oai_token-act-pair · gpt-3.5-turbo

instructions or calls to action

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

 sadd

-0.69

 reneg

-0.68

 presided

-0.68

 upsetting

-0.66

 reunited

-0.66

 supposedly

-0.66

 certainly

-0.65

 indeed

-0.65

 reunion

-0.65

quo

-0.64

POSITIVE LOGITS

Use

3.31

Use

2.54

 Uses

2.20

USE

2.07

use

1.81

 Usage

1.76

use

1.73

 Using

1.69

 Used

1.54

Using

1.43

Activations Density 0.012%

phrases indicating instructions or recommendations

instructions or calls to action

No Comments

No Known Activations