INDEX

Explanations

instances where something is given a new or different name, often for clarity or emphasis

oai_token-act-pair · gpt-3.5-turbo

instances of the word "dubbed" in various contexts

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

emic

-0.68

abet

-0.67

isman

-0.65

acion

-0.65

alg

-0.65

ramid

-0.64

issue

-0.63

idem

-0.63

isner

-0.63

 Admin

-0.61

POSITIVE LOGITS

 dubbed

1.27

 nicknamed

1.03

named

0.76

istani

0.71

 termed

0.70

nik

0.70

forth

0.69

geon

0.69

INGTON

0.69

 labelled

0.68

Activations Density 0.008%

instances where something is given a new or different name, often for clarity or emphasis

instances of the word "dubbed" in various contexts

No Comments

No Known Activations