INDEX

Explanations

instances where the word "hi" is present with a high activation value, potentially indicating a specific focus on this word

oai_token-act-pair · gpt-3.5-turbo

repetitions of the phrase "hi."

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

 Cosponsors

-0.77

 Aven

-0.72

 Sorceress

-0.68

 Izan

-0.65

 convol

-0.64

 Jenner

-0.64

ilater

-0.62

 Euph

-0.62

orative

-0.62

NetMessage

-0.60

POSITIVE LOGITS

hi

1.16

ya

1.07

emen

0.98

hei

0.95

roth

0.93

oga

0.90

emi

0.89

wa

0.89

omo

0.87

oka

0.87

Activations Density 0.006%

instances where the word "hi" is present with a high activation value, potentially indicating a specific focus on this word

repetitions of the phrase "hi."

No Comments

No Known Activations