INDEX

Explanations

phrases related to news articles or features marked as "Full" or "Fuller."

oai_token-act-pair · gpt-3.5-turbo

instances of the word "Full."

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

 chill

-0.66

 harb

-0.65

 thought

-0.65

 wounds

-0.65

 imper

-0.63

 memor

-0.63

inc

-0.62

 wound

-0.62

 stereotyp

-0.60

 nerves

-0.60

POSITIVE LOGITS

Full

3.71

full

2.32

 Full

2.11

 FULL

1.71

 full

1.64

 Fuller

1.55

 fuller

1.41

Half

1.39

partial

1.21

 fullest

1.21

Activations Density 0.013%

phrases related to news articles or features marked as "Full" or "Fuller."

instances of the word "Full."

No Comments

No Known Activations