INDEX

Explanations

terms related to being user-friendly or suitable for a specific audience

oai_token-act-pair · gpt-3.5-turbo

terms related to friendliness

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 outl

-0.66

IV

-0.64

ax

-0.63

 principal

-0.63

 subdiv

-0.63

 âĢ

-0.63

 [...]

-0.62

 marrow

-0.62

ded

-0.61

 transc

-0.61

POSITIVE LOGITS

friendly

3.86

 Friendly

2.18

 friendly

2.10

Friend

1.99

friend

1.59

safe

1.39

riend

1.27

loving

1.27

oriented

1.25

happy

1.24

Activations Density 0.022%

terms related to being user-friendly or suitable for a specific audience

terms related to friendliness

No Comments

No Known Activations