INDEX

Explanations

the word "be" at a high activation level

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_fs12288-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_12288

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

12,288

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 Stain

-0.68

andra

-0.60

 Dresden

-0.59

 Moose

-0.59

 Corpus

-0.58

 Prescott

-0.57

 Dude

-0.56

rament

-0.56

 Malk

-0.56

achine

-0.56

POSITIVE LOGITS

 able

0.88

 sure

0.86

 honest

0.85

 replaced

0.85

 considered

0.85

 seen

0.82

heading

0.81

 frank

0.79

fits

0.78

 Continued

0.76

Activations Density 0.095%

the word "be" at a high activation level

No Comments

No Known Activations

the word "be" at a high activation level

No Comments

No Known Activations