INDEX

Explanations

formal communication fragments, including names and titles

oai_token-act-pair · gpt-3.5-turbo

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

oiler

-0.77

 gore

-0.73

 dunno

-0.70

 hardcore

-0.69

 legit

-0.68

 partying

-0.68

 Panic

-0.68

 cops

-0.68

umat

-0.68

 destro

-0.68

POSITIVE LOGITS

 Honour

1.46

sir

1.28

 gentlemen

1.24

 respectfully

1.17

 gentleman

1.14

 Chairman

1.13

Chair

1.11

 Thank

1.06

reetings

1.06

 Gentle

1.06

Activations Density 0.671%

formal communication fragments, including names and titles

No Comments

No Known Activations