INDEX

Explanations

proper names, specifically "Henry"

oai_token-act-pair · gpt-3.5-turbo

references to the name "Henry."

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

 discriminated

-0.79

 âĶ

-0.71

 bases

-0.69

 polar

-0.68

utt

-0.67

 nuts

-0.66

 plane

-0.64

 prevail

-0.64

 nets

-0.62

 Spac

-0.62

POSITIVE LOGITS

 Henry

3.46

Henry

2.95

 Edward

1.51

 Kissinger

1.38

 Richard

1.37

 Henri

1.34

 Wilhelm

1.29

 Charles

1.27

 Arthur

1.27

 Philip

1.26

Activations Density 0.019%

proper names, specifically "Henry"

references to the name "Henry."

No Comments

No Known Activations