INDEX

Explanations

long sequences of capital letters

oai_token-act-pair · gpt-3.5-turbo

repeated mentions of the word "Long."

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

 babys

-0.68

 seating

-0.67

 acting

-0.67

 cabinet

-0.66

 applicable

-0.65

 trust

-0.64

 recept

-0.63

 availability

-0.62

 attending

-0.62

 Arch

-0.62

POSITIVE LOGITS

Long

3.59

long

2.07

Short

1.96

 LONG

1.69

 Long

1.65

Little

1.41

short

1.33

Large

1.30

Old

1.29

Length

1.28

Activations Density 0.016%

long sequences of capital letters

repeated mentions of the word "Long."

No Comments

No Known Activations