INDEX

Explanations

the word "Mars" with varying activations

oai_token-act-pair · gpt-3.5-turbo

mentions of the planet Mars

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

talk

-0.77

)</

-0.71

 trophies

-0.66

voice

-0.66

 purse

-0.65

FN

-0.64

FN

-0.63

 intrins

-0.62

oho

-0.62

ND

-0.61

POSITIVE LOGITS

 Mars

3.89

Mars

3.42

 Martian

2.16

 Venus

1.97

 Ceres

1.82

 Pluto

1.74

 rover

1.70

 mars

1.68

 Jupiter

1.49

 Mercury

1.43

Activations Density 0.016%

the word "Mars" with varying activations

mentions of the planet Mars

No Comments

No Known Activations