INDEX

Explanations

words related to being the recipient or target of an action or concept

oai_token-act-pair · gpt-3.5-turbo

words and phrases related to forms of the word "ject" and its variants

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

Negative Logits

 timber

-0.69

 covered

-0.67

 rose

-0.63

gig

-0.61

 Artemis

-0.60

 bandits

-0.59

 Zhou

-0.59

Ron

-0.59

 Santa

-0.58

 Barth

-0.58

POSITIVE LOGITS

ject

4.35

jection

3.20

jected

2.73

JECT

1.93

 inject

1.28

herent

1.20

fect

1.14

cess

1.12

rupt

1.06

spect

1.05

Activations Density 0.018%

words related to being the recipient or target of an action or concept

words and phrases related to forms of the word "ject" and its variants

No Known Activations