INDEX

Explanations

forms of emphasis, such as exclamation marks or capitalized words, which indicate strong emotions or stress

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

%%%%

-0.73

@@

-0.65

 Scrolls

-0.64

 Theft

-0.59

quickShipAvailable

-0.59

 signs

-0.56

 smokes

-0.56

 splits

-0.55

 crept

-0.55

missing

-0.55

POSITIVE LOGITS

venge

1.37

 extent

1.36

 detriment

1.29

 tune

1.13

rouse

1.12

livion

1.03

 shores

0.97

 fullest

0.95

 liking

0.90

 brink

0.88

Activations Density 4.801%

forms of emphasis, such as exclamation marks or capitalized words, which indicate strong emotions or stress

No Comments

No Known Activations