INDEX

Explanations

names of locations and possibly names of people

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

FORMATION

-0.76

 crank

-0.76

 fertil

-0.72

 depreciation

-0.71

 sight

-0.71

CLASSIFIED

-0.70

 dividing

-0.66

 checklist

-0.65

 lounge

-0.64

 gravity

-0.64

POSITIVE LOGITS

ona

1.18

orius

1.12

inda

1.09

aya

1.09

arus

1.09

ira

1.08

aka

1.08

inia

1.07

rik

1.07

ani

1.07

Activations Density 3.292%

names of locations and possibly names of people

No Comments

No Known Activations