INDEX

Explanations

tokens representing a specific code or abbreviation 'PU' followed by a number

oai_token-act-pair · gpt-3.5-turbo

references to publishing and publication-related terms

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

 Jenkins

-0.70

 Heath

-0.70

alia

-0.67

 cruise

-0.66

 Howell

-0.65

 Helm

-0.65

 Hart

-0.64

 chron

-0.64

 Lamp

-0.64

rica

-0.62

POSITIVE LOGITS

PU

4.41

PU

1.93

pu

1.40

MU

1.35

PLA

1.18

SU

1.16

PG

1.13

PE

1.13

PUT

1.12

BU

1.11

Activations Density 0.022%

tokens representing a specific code or abbreviation 'PU' followed by a number

references to publishing and publication-related terms

No Comments

No Known Activations