INDEX

Explanations

expressions indicating a need for a certain card to not be duplicated in a deck

oai_token-act-pair · gpt-3.5-turbo

instances of the word "no."

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

RAFT

-0.81

iership

-0.79

rious

-0.70

rosse

-0.69

minster

-0.68

lycer

-0.67

romy

-0.62

lish

-0.62

bledon

-0.61

realDonaldTrump

-0.61

POSITIVE LOGITS

xious

1.27

 longer

0.98

zzle

0.96

ises

0.94

obs

0.93

ct

0.92

ise

0.90

oses

0.89

 matter

0.88

vell

0.86

Activations Density 0.075%

expressions indicating a need for a certain card to not be duplicated in a deck

instances of the word "no."

No Comments

No Known Activations