INDEX

Explanations

highly emphasized words related to important or significant topics

oai_token-act-pair · gpt-3.5-turbo

references to significant concepts or issues, indicating a focus on underlying themes within narratives

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

 Carbuncle

-0.73

rily

-0.72

uly

-0.71

cknowled

-0.68

uled

-0.68

region

-0.66

viously

-0.65

ega

-0.65

cellaneous

-0.61

pecially

-0.60

POSITIVE LOGITS

DonaldTrump

0.80

hyde

0.70

ICAN

0.69

UID

0.67

acea

0.67

behind

0.65

×Ļ×

0.65

wives

0.64

ument

0.64

ãĤ·ãĥ£

0.63

Activations Density 0.219%

highly emphasized words related to important or significant topics

references to significant concepts or issues, indicating a focus on underlying themes within narratives

No Comments

No Known Activations