INDEX

Explanations

words related to leaked or unauthorized information/documents

oai_token-act-pair · gpt-3.5-turbo

references to leaked information or documents

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

rior

-0.71

Sax

-0.68

 Atkins

-0.61

 Ages

-0.61

joy

-0.59

.}

-0.59

orest

-0.59

rest

-0.59

ians

-0.59

 orient

-0.58

POSITIVE LOGITS

 leaked

3.85

 leaks

2.56

 leak

2.39

 leaking

2.15

Leaks

1.66

 leakage

1.56

 spilled

1.55

 Wikileaks

1.49

 circulated

1.45

 hacked

1.43

Activations Density 0.012%

words related to leaked or unauthorized information/documents

references to leaked information or documents

No Comments

No Known Activations