INDEX

Explanations

instances where someone did not directly respond or provide a specific answer

oai_token-act-pair · gpt-3.5-turbo

negative statements and denials

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

 Compass

-0.75

 Kinnikuman

-0.73

 Creat

-0.66

ranged

-0.64

EStream

-0.63

Pop

-0.62

 Empires

-0.60

Intern

-0.60

 Roses

-0.59

 Reborn

-0.59

POSITIVE LOGITS

 specify

1.30

 hesitate

1.26

 mention

1.21

 elaborate

1.17

 disclose

1.16

 deny

1.08

 delve

1.04

 explicitly

1.03

 clarify

1.02

 cite

1.02

Activations Density 0.087%

instances where someone did not directly respond or provide a specific answer

negative statements and denials

No Comments

No Known Activations