INDEX

Explanations

mentions of high value or high quality items

oai_token-act-pair · gpt-3.5-turbo

instances of the word "High" as well as related high-value mentions

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

 nucleus

-0.74

 inclusive

-0.63

atorial

-0.63

elia

-0.63

 recovery

-0.63

 sweater

-0.61

 breathing

-0.61

 spokesman

-0.61

 voic

-0.60

arium

-0.60

POSITIVE LOGITS

High

3.17

Low

1.83

Hig

1.54

Higher

1.52

high

1.41

Extreme

1.37

Poor

1.29

Lower

1.29

Top

1.23

Medium

1.22

Activations Density 0.017%

mentions of high value or high quality items

instances of the word "High" as well as related high-value mentions

No Comments

No Known Activations