INDEX

Explanations

requests or statements followed by a specific action or instruction

oai_token-act-pair · gpt-3.5-turbo

instances of the word "Please," indicating a request or instruction

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

ounter

-0.70

 attributes

-0.64

 early

-0.64

 shifting

-0.64

 secret

-0.63

 diss

-0.63

 marvel

-0.62

 prototyp

-0.62

 skirm

-0.62

 latent

-0.61

POSITIVE LOGITS

 Please

3.25

Please

2.34

 PLEASE

2.26

please

2.13

 please

2.05

 Sorry

1.43

 Thank

1.41

PLE

1.25

 Feel

1.24

 Therefore

1.22

Activations Density 0.023%

requests or statements followed by a specific action or instruction

instances of the word "Please," indicating a request or instruction

No Comments

No Known Activations