INDEX

Explanations

instances where something is not necessarily true or accurate

oai_token-act-pair · gpt-3.5-turbo

phrases that emphasize the idea of being non-compulsory or conditional in nature

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

sel

-0.75

DL

-0.73

tein

-0.69

banks

-0.66

ories

-0.66

 Tate

-0.65

EE

-0.65

 Marian

-0.65

sche

-0.62

ZE

-0.62

POSITIVE LOGITS

 necessarily

1.12

 necess

1.06

entimes

0.97

necess

0.94

 detract

0.93

 guiActiveUn

0.90

wcsstore

0.86

 obliged

0.84

 exha

0.83

 cumbers

0.83

Activations Density 0.005%

instances where something is not necessarily true or accurate

phrases that emphasize the idea of being non-compulsory or conditional in nature

No Comments

No Known Activations