INDEX

Explanations

phrases related to counting or quantifying

oai_token-act-pair · gpt-3.5-turbo

references to audiences or demographics in discussions

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

 sunset

-0.59

 Collider

-0.56

 Diver

-0.54

 Versus

-0.54

 Simulation

-0.53

 Moonlight

-0.52

 Chimera

-0.51

 Vulkan

-0.51

 Lethal

-0.51

 Archdemon

-0.50

POSITIVE LOGITS

selves

0.69

 whom

0.68

who

0.64

heastern

0.64

rosse

0.63

dinand

0.63

anners

0.62

irs

0.61

 alike

0.61

hari

0.59

Activations Density 1.944%

phrases related to counting or quantifying

references to audiences or demographics in discussions

No Comments

No Known Activations