INDEX

Explanations

terms related to news outlets such as NPR

oai_token-act-pair · gpt-3.5-turbo

references to NPR and PBS

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

loader

-0.68

ãĤ¹ãĥĪ

-0.67

gart

-0.66

jamin

-0.66

wagen

-0.64

oldemort

-0.63

folios

-0.63

uay

-0.63

 Emirates

-0.62

 Painter

-0.61

POSITIVE LOGITS

NPR

1.17

NPR

1.04

DEF

0.84

REC

0.83

 transcripts

0.79

DAQ

0.76

ICA

0.74

 Transcript

0.74

TIT

0.74

IC

0.74

Activations Density 0.008%

terms related to news outlets such as NPR

references to NPR and PBS

No Comments

No Known Activations