INDEX

Explanations

years expressed in the format "U.S." followed by a number

oai_token-act-pair · gpt-3.5-turbo

references to the U.S. government and its associated statistics and reports

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 10-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.10.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.10.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

azaar

-0.72

ebook

-0.67

ructure

-0.66

hid

-0.64

veyard

-0.64

zzle

-0.63

alys

-0.63

ription

-0.62

lass

-0.62

aird

-0.62

POSITIVE LOGITS

 Airways

1.08

 taxpayers

1.05

 Airlines

1.04

 officials

1.00

 embassies

0.98

 diplomats

0.98

 lawmakers

0.98

 presidents

0.96

 soil

0.94

 citizens

0.94

Activations Density 0.088%

years expressed in the format "U.S." followed by a number

references to the U.S. government and its associated statistics and reports

No Comments

No Known Activations