INDEX

Explanations

specific mention of the word "bullet" at a high activation level

oai_token-act-pair · gpt-3.5-turbo

references to bullets or bullet points in various contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

Sensor

-0.76

Suc

-0.73

 Mistress

-0.72

wcsstore

-0.71

CRE

-0.68

Place

-0.67

 Valhalla

-0.65

UTC

-0.65

kef

-0.64

ËĪ

-0.64

POSITIVE LOGITS

proof

1.14

 bullet

1.05

 bullets

0.94

 Bullet

0.85

 holes

0.84

 hole

0.78

 hoops

0.78

heads

0.75

 dodged

0.75

atical

0.73

Activations Density 0.012%

specific mention of the word "bullet" at a high activation level

references to bullets or bullet points in various contexts

No Comments

No Known Activations