INDEX

Explanations

commands or instructions typically ending with a verb and a dash

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

 Arabian

-0.64

 harbour

-0.62

 hiber

-0.62

 flirt

-0.62

 fright

-0.60

 caution

-0.59

 track

-0.59

 lull

-0.59

 polyg

-0.59

 subdiv

-0.58

POSITIVE LOGITS

turned

1.39

cum

1.21

selves

1.19

sama

1.18

style

1.14

related

1.13

sized

1.13

type

1.12

looking

1.07

themed

1.06

Activations Density 0.123%

commands or instructions typically ending with a verb and a dash

No Comments

No Known Activations