INDEX

Explanations

phrases that emphasize instructions and actions

oai_token-act-pair · gpt-3.5-turbo

contrasting phrases and qualifiers in the text

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

lde

-0.74

 Suddenly

-0.72

Suddenly

-0.67

 ãĤµãĥ¼ãĥĨãĤ£ãĥ¯ãĥ³

-0.66

literally

-0.66

Huh

-0.66

bang

-0.65

And

-0.65

irit

-0.64

hab

-0.62

POSITIVE LOGITS

 please

1.35

please

1.09

 refunds

1.07

 PLEASE

1.02

Please

0.92

 Please

0.91

 exceptions

0.90

you

0.87

 beware

0.86

 discretion

0.85

Activations Density 0.369%

phrases that emphasize instructions and actions

contrasting phrases and qualifiers in the text

No Comments

No Known Activations