INDEX

Explanations

phrases related to completion or improvement of tasks or actions

oai_token-act-pair · gpt-3.5-turbo

instances of the word "up."

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

ctions

-0.73

vernment

-0.73

ciation

-0.68

Org

-0.67

ionics

-0.66

rang

-0.66

rand

-0.61

ident

-0.61

ITNESS

-0.57

nor

-0.57

POSITIVE LOGITS

up

2.90

Up

2.05

up

2.03

Up

1.88

UP

1.81

ups

1.68

ups

1.59

UP

1.54

upt

1.47

 upwards

1.42

Activations Density 0.146%

phrases related to completion or improvement of tasks or actions

instances of the word "up."

No Comments

No Known Activations