INDEX

Explanations

the word "couldn't" with high activation values

oai_token-act-pair · gpt-3.5-turbo

instances of the word "couldn't."

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

oak

-0.69

dress

-0.67

 protected

-0.65

 liberated

-0.62

backer

-0.62

croft

-0.61

ULT

-0.60

PU

-0.60

 deposition

-0.58

otype

-0.58

POSITIVE LOGITS

't

1.51

adian

0.96

 afford

0.91

atio

0.90

kered

0.90

ayan

0.85

anke

0.84

 feas

0.84

ilater

0.80

ieve

0.79

Activations Density 0.012%

the word "couldn't" with high activation values

instances of the word "couldn't."

No Comments

No Known Activations