INDEX

Explanations

references to actions that are being redone or repeated

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

DragonMagazine

-0.77

BuyableInstoreAndOnline

-0.74

 Pebble

-0.71

 Nest

-0.65

 buckle

-0.62

 Sammy

-0.62

 Northern

-0.59

 Sturgeon

-0.58

dor

-0.58

 Indonesian

-0.57

POSITIVE LOGITS

becca

1.19

versible

1.09

brand

1.08

jected

1.08

porter

1.07

interpret

1.07

venge

1.05

usable

1.04

ordering

1.04

levant

1.03

Activations Density 2.079%

references to actions that are being redone or repeated

No Comments

No Known Activations