INDEX

Explanations

people's names with the specific pattern of being followed by a single capital letter and a non-zero activation value

oai_token-act-pair · gpt-3.5-turbo

references to individuals or names

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

IFrame

Link

Negative Logits

tem

-0.66

 reciproc

-0.64

 subsid

-0.62

 nonex

-0.61

 bearing

-0.60

 disparate

-0.60

rud

-0.60

 skate

-0.60

 disappro

-0.59

 contradict

-0.58

POSITIVE LOGITS

eely

1.33

aylor

1.25

itsch

1.17

unn

1.17

ieves

1.16

evin

1.11

ettle

1.10

asser

1.09

elsen

1.08

olen

1.08

Activations Density 0.024%

people's names with the specific pattern of being followed by a single capital letter and a non-zero activation value

references to individuals or names

No Known Activations