INDEX

Explanations

names of political figures, government positions, and organizations

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

reci

-0.80

 relate

-0.74

fit

-0.73

 grate

-0.72

 inval

-0.72

 cords

-0.72

 surprises

-0.72

 needles

-0.71

headers

-0.71

ean

-0.70

POSITIVE LOGITS

 Carnegie

1.09

 University

1.05

Nex

1.01

 National

1.01

 Univ

0.99

 Georgetown

0.99

 Harvard

0.98

 United

0.98

 Fairfax

0.96

 Columbia

0.95

Activations Density 6.441%

names of political figures, government positions, and organizations

No Comments

No Known Activations