INDEX

Explanations

news-related words and phrases

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

cially

-0.84

Recommend

-0.83

Additionally

-0.81

amount

-0.79

METHOD

-0.78

Import

-0.77

Further

-0.77

specified

-0.77

External

-0.77

Internal

-0.76

POSITIVE LOGITS

nap

1.11

 gigg

1.09

 sweaty

1.08

 humming

1.07

 candy

1.06

 grandma

1.06

 vomit

1.03

 pancakes

1.02

 stuffed

1.02

 popcorn

1.02

Activations Density 11.004%

news-related words and phrases

No Comments

No Known Activations