INDEX

Explanations

ways

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_fs3072-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_3072

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

3,072

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

zona

-0.77

 Noir

-0.62

Â¯Â¯Â¯Â¯

-0.61

isure

-0.58

Constructed

-0.56

 lobe

-0.56

 membr

-0.56

oning

-0.56

 Reborn

-0.55

phone

-0.55

POSITIVE LOGITS

geon

1.23

ierrez

1.18

geons

1.13

geoning

1.06

ulkan

0.99

apest

0.99

culosis

0.98

rition

0.96

gery

0.94

uru

0.94

Activations Density 1.350%

ways

No Comments

No Known Activations