INDEX

Explanations

numbers, specifically the number "ten" or close variations

oai_token-act-pair · gpt-3.5-turbo

instances of the word "ten" or references related to the number ten

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

aker

-0.73

jri

-0.70

ault

-0.70

asks

-0.68

Adds

-0.65

poke

-0.64

cation

-0.63

MQ

-0.63

 Jackets

-0.63

ibles

-0.63

POSITIVE LOGITS

ten

3.11

 fifteen

2.30

 twenty

2.19

 twelve

2.19

 eleven

2.16

 fourteen

2.10

 thirty

2.08

 thirteen

2.08

 sixteen

2.05

 fifty

2.02

Activations Density 0.011%

numbers, specifically the number "ten" or close variations

instances of the word "ten" or references related to the number ten

No Comments

No Known Activations