INDEX

Explanations

political terms and names

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

 decomp

-0.80

 metab

-0.74

 explorers

-0.74

 uploading

-0.73

 catalogue

-0.73

atos

-0.70

 Grimoire

-0.70

 experien

-0.70

 overload

-0.70

 productions

-0.67

POSITIVE LOGITS

Senate

1.60

Republican

1.55

Republicans

1.54

 Senate

1.53

 reelection

1.51

 Republican

1.51

Democrats

1.48

 Democrat

1.44

 caucus

1.41

 senate

1.39

Activations Density 7.555%

political terms and names

No Comments

No Known Activations