INDEX

Explanations

references to physical or metaphorical defects and errors

oai_token-act-pair · gpt-3.5-turbo

references to various types of defects

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

 Pant

-0.75

 Hawaiian

-0.72

Wet

-0.72

Music

-0.71

HER

-0.69

Las

-0.68

 Hearthstone

-0.67

 Hawai

-0.66

 Cena

-0.66

 Mult

-0.64

POSITIVE LOGITS

 defect

4.54

 defects

2.84

 defective

2.00

 flaw

1.47

 deficiency

1.41

 malfunction

1.25

 faulty

1.22

 fault

1.13

 defic

1.10

 deform

1.09

Activations Density 0.020%

references to physical or metaphorical defects and errors

references to various types of defects

No Comments

No Known Activations