INDEX

Explanations

instances of the word "mal" indicating malfeasance or misconduct

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs49152-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_49152

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

49,152

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

 Carbuncle

-0.74

 Fargo

-0.72

æĸ¹

-0.68

 Salvation

-0.67

 Wide

-0.66

 Defenders

-0.66

ALK

-0.65

 Polk

-0.64

 Decoder

-0.64

 Hobby

-0.63

POSITIVE LOGITS

ignant

1.32

practice

1.18

colm

1.15

absor

1.15

adies

1.13

formed

1.10

igned

1.06

igning

1.02

ady

1.00

adjusted

1.00

Activations Density 0.005%

instances of the word "mal" indicating malfeasance or misconduct

No Comments

No Known Activations