INDEX

Explanations

interjections and verbal fillers.Significant activation was also seen for medical conditions

oai_token-act-pair · gpt-3.5-turbo

interjections, particularly variations of "uh" and "eh."

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

å¥³

-0.86

IUM

-0.81

xual

-0.80

BOOK

-0.79

flies

-0.75

NetMessage

-0.73

krit

-0.73

ioned

-0.73

milo

-0.71

SEA

-0.69

POSITIVE LOGITS

 pardon

0.81

Huh

0.79

oh

0.79

huh

0.77

Uh

0.75

bye

0.73

uh

0.72

rah

0.72

awk

0.70

ahah

0.69

Activations Density 0.004%

interjections and verbal fillers.Significant activation was also seen for medical conditions

interjections, particularly variations of "uh" and "eh."

No Comments

No Known Activations