INDEX

Explanations

phrases related to tasks being completed or finished

oai_token-act-pair · gpt-3.5-turbo

instances of the word "done."

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

Å¡

-0.70

anth

-0.65

lement

-0.62

unin

-0.61

anta

-0.59

 McCorm

-0.58

 correspond

-0.56

Ann

-0.56

acas

-0.54

erald

-0.53

POSITIVE LOGITS

 done

3.60

done

2.82

 Done

2.29

Done

2.08

 accomplished

1.77

 undertaken

1.72

 performed

1.68

 finished

1.39

 completed

1.38

 achieved

1.34

Activations Density 0.021%

phrases related to tasks being completed or finished

instances of the word "done."

No Comments

No Known Activations