INDEX

Explanations

references to specific TV channels followed by numbers

oai_token-act-pair · gpt-3.5-turbo

references to television channels

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

oren

-0.66

ger

-0.66

 advoc

-0.65

 lettuce

-0.64

redo

-0.64

orer

-0.63

 illeg

-0.62

 relational

-0.61

ol

-0.60

osaurs

-0.60

POSITIVE LOGITS

 Channel

4.22

Channel

3.03

 channel

2.29

channel

2.03

 channels

1.89

annels

1.54

ITV

1.25

 Broadcasting

1.24

 Radio

1.10

 Network

1.09

Activations Density 0.007%

references to specific TV channels followed by numbers

references to television channels

No Comments

No Known Activations