INDEX

Explanations

mentions of historical or practical information in a compressed format

oai_token-act-pair · gpt-3.5-turbo

references to specific individuals, concepts regarding economic or health contexts, and phrases that summarize information or indicate avoidance behavior

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

 Kard

-0.77

 Explos

-0.67

 embell

-0.64

 Pend

-0.61

 Prism

-0.61

 entrants

-0.60

Lav

-0.58

 kilomet

-0.57

 sides

-0.57

 upload

-0.56

POSITIVE LOGITS

uchin

3.18

 nutshell

1.82

 avoidance

1.28

utch

1.20

hay

1.14

uty

1.13

utive

1.01

jen

1.00

rawdownloadcloneembedreportprint

0.95

 avoiding

0.93

Activations Density 0.031%

mentions of historical or practical information in a compressed format

references to specific individuals, concepts regarding economic or health contexts, and phrases that summarize information or indicate avoidance behavior

No Comments

No Known Activations