INDEX

Explanations

phrases indicating someone has something to say or share

oai_token-act-pair · gpt-3.5-turbo

phrases that express the act of conveying or reporting information

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

riger

-0.69

hower

-0.69

nces

-0.68

croft

-0.67

irection

-0.65

 Continued

-0.65

 simultane

-0.65

ihad

-0.63

 horizont

-0.60

cephal

-0.60

POSITIVE LOGITS

 offer

1.25

 contribute

0.97

say

0.96

 lose

0.94

 prove

0.93

 gain

0.91

SAY

0.90

 contend

0.90

 impart

0.85

 apologize

0.85

Activations Density 0.042%

phrases indicating someone has something to say or share

phrases that express the act of conveying or reporting information

No Comments

No Known Activations