INDEX

Explanations

words related to legal and investigative matters, such as evidence, documents, information, and investigations

oai_token-act-pair · gpt-3.5-turbo

references to documents or evidence related to investigations or legal cases

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.11.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

 preferring

-0.63

 Zombies

-0.62

 substit

-0.61

 Punch

-0.60

ovy

-0.60

 rite

-0.60

ointed

-0.59

think

-0.58

 hating

-0.58

 masc

-0.57

POSITIVE LOGITS

 incrim

1.41

 corrobor

1.19

 reveal

1.15

 detailing

1.02

 exoner

1.02

 implicated

1.00

 contain

1.00

 detail

0.99

 discl

0.99

 divul

0.99

Activations Density 0.303%

words related to legal and investigative matters, such as evidence, documents, information, and investigations

references to documents or evidence related to investigations or legal cases

No Comments

No Known Activations