INDEX

Explanations

confession-related phrases, such as "confessing," "confessed," and "confession."

oai_token-act-pair · gpt-3.5-turbo Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs768-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_768

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

768

Data Type

torch.float32

Hook Point

blocks.8.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

hyde

-0.82

VILLE

-0.76

 Elves

-0.74

WOOD

-0.71

OHN

-0.71

 horizontally

-0.69

tone

-0.68

hunter

-0.67

DAY

-0.67

 Bard

-0.67

POSITIVE LOGITS

ention

1.41

icit

1.39

icted

1.39

orted

1.38

ervation

1.31

idential

1.31

osed

1.30

iction

1.30

ortion

1.28

inent

1.26

Activations Density 2.822%

confession-related phrases, such as "confessing," "confessed," and "confession."

No Comments

No Known Activations