INDEX

Explanations

contractions of words, specifically finding instances of "didn't" with a strong activation value

oai_token-act-pair · gpt-3.5-turbo

negative contractions and phrases that express negation

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

planet

-0.75

amer

-0.71

 rall

-0.64

accompan

-0.64

bard

-0.63

stre

-0.62

 Britann

-0.62

antine

-0.62

rog

-0.61

Reviewer

-0.60

POSITIVE LOGITS

 necessarily

1.03

 exactly

1.02

 gonna

0.95

 quite

0.84

urtles

0.82

 gotta

0.81

 kidding

0.78

 really

0.77

 bother

0.76

 even

0.75

Activations Density 0.069%

contractions of words, specifically finding instances of "didn't" with a strong activation value

negative contractions and phrases that express negation

No Comments

No Known Activations