INDEX

Explanations

verbs in the negative form with anticipation of what is to come next

oai_token-act-pair · gpt-3.5-turbo

instances of the word "won."

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

amaz

-0.76

aq

-0.72

ibly

-0.63

acting

-0.62

ipes

-0.62

Subject

-0.61

ali

-0.60

atively

-0.60

 Ster

-0.59

reference

-0.58

POSITIVE LOGITS

won

3.38

won

2.43

Won

2.24

 wont

1.63

 wouldn

1.36

 shouldn

1.34

 wins

1.31

'll

1.22

 will

1.16

 winning

1.13

Activations Density 0.034%

verbs in the negative form with anticipation of what is to come next

instances of the word "won."

No Comments

No Known Activations