INDEX

Explanations

words related to news or announcements

oai_token-act-pair · gpt-3.5-turbo

references to a specific entity or organization, likely related to the "UE" acronym

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

 Harris

-0.74

pher

-0.72

lob

-0.66

Har

-0.65

 babys

-0.64

plan

-0.64

 Mitt

-0.62

har

-0.61

 defense

-0.60

 Hook

-0.59

POSITIVE LOGITS

UE

4.68

UES

2.92

ues

2.41

ue

2.03

ued

2.02

uing

1.82

UI

1.54

uers

1.50

uer

1.39

UME

1.36

Activations Density 0.010%

words related to news or announcements

references to a specific entity or organization, likely related to the "UE" acronym

No Comments

No Known Activations