INDEX

Explanations

Twitter usernames

oai_token-act-pair · gpt-3.5-turbo

numeric sequences and identifiers, likely related to URLs or social media handles

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.8.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

 stuffing

-0.66

 fools

-0.65

 announ

-0.64

asts

-0.64

 livest

-0.62

 fundra

-0.62

 lect

-0.61

 thirds

-0.61

 distractions

-0.61

 losers

-0.59

POSITIVE LOGITS

bj

0.95

zx

0.91

OX

0.90

0.88

df

0.87

Eh

0.87

xus

0.86

CN

0.85

sg

0.85

wm

0.85

Activations Density 0.040%

Twitter usernames

numeric sequences and identifiers, likely related to URLs or social media handles

No Comments

No Known Activations