INDEX

Explanations

capitalized abbreviations with numbers at the end, possibly related to technical terms or codes

oai_token-act-pair · gpt-3.5-turbo

references to "ST" or related terms indicating a standard or systems terminology

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

Mai

-0.75

omever

-0.72

ocument

-0.69

javascript

-0.67

âĢİ

-0.66

zai

-0.66

ajo

-0.64

cia

-0.63

 compr

-0.61

dyl

-0.60

POSITIVE LOGITS

ST

3.56

ST

1.87

STD

1.69

STA

1.50

STE

1.37

STR

1.35

ART

1.34

STL

1.31

St

1.29

SC

1.29

Activations Density 0.012%

capitalized abbreviations with numbers at the end, possibly related to technical terms or codes

references to "ST" or related terms indicating a standard or systems terminology

No Comments

No Known Activations