INDEX

Explanations

phrases related to command or authority

oai_token-act-pair · gpt-3.5-turbo

references to control or authority within a context

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

 TOUR

-0.72

INGTON

-0.68

PET

-0.67

 execut

-0.67

 voic

-0.65

 Exile

-0.65

iP

-0.64

 chars

-0.62

 Atomic

-0.62

 Bund

-0.62

POSITIVE LOGITS

acea

0.82

ibaba

0.81

ndra

0.81

osate

0.75

prus

0.74

oglu

0.73

ossier

0.73

lda

0.71

aund

0.69

hai

0.69

Activations Density 0.000%

phrases related to command or authority

references to control or authority within a context

No Comments

No Known Activations