INDEX

Explanations

the word "mask" with high activations, and related words like "disguise" with lower activations

oai_token-act-pair · gpt-3.5-turbo

references to masks and disguises

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

athan

-0.74

âĶĢâĶĢâĶĢâĶĢâĶĢâĶĢâĶĢâĶĢ

-0.72

scill

-0.72

course

-0.70

GGGGGGGG

-0.68

 Yards

-0.66

lished

-0.66

ALLY

-0.66

ally

-0.66

rian

-0.65

POSITIVE LOGITS

 masks

1.12

 mask

1.06

resses

0.91

 Mask

0.88

mask

0.83

 wearer

0.80

Mask

0.80

 worn

0.80

 concealed

0.79

ħĭ

0.77

Activations Density 0.017%

the word "mask" with high activations, and related words like "disguise" with lower activations

references to masks and disguises

No Comments

No Known Activations