INDEX

Explanations

names of buildings or historical places

oai_token-act-pair · gpt-3.5-turbo

references to specific palaces or significant historical locations

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 4-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.4.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.4.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

CAST

-0.73

ELL

-0.70

VOL

-0.70

::::::::

-0.69

err

-0.67

ivity

-0.67

arette

-0.66

REE

-0.65

CHAT

-0.64

 Braun

-0.64

POSITIVE LOGITS

 Palace

1.00

ibur

0.86

osaurs

0.81

 palace

0.80

gur

0.76

 maiden

0.76

monary

0.75

itiz

0.75

 Hotel

0.74

OTUS

0.72

Activations Density 0.011%

names of buildings or historical places

references to specific palaces or significant historical locations

No Comments

No Known Activations