INDEX

Explanations

names of publications, particularly "The Wall Street Journal"

oai_token-act-pair · gpt-3.5-turbo

references to the Wall Street Journal

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

Not in Any Lists

No Comments

Negative Logits

netflix

-0.85

ibilities

-0.77

essa

-0.74

abase

-0.71

eworld

-0.70

quit

-0.69

lain

-0.67

combe

-0.67

orest

-0.66

win

-0.65

POSITIVE LOGITS

 Journal

1.45

 regulator

0.85

Journal

0.85

rollers

0.84

 firms

0.83

 bankers

0.81

 analysts

0.81

 banks

0.79

 banker

0.78

 Analyst

0.78

Activations Density 0.039%

names of publications, particularly "The Wall Street Journal"

references to the Wall Street Journal

No Comments

No Known Activations