INDEX

Explanations

words related to expressions such as "Well," or "Do," that are commonly used at the beginning of sentences for emphasis or to introduce a new idea

oai_token-act-pair · gpt-3.5-turbo

conversational prompts or interjections that initiate statements or questions

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

**

-0.89

âĢķ

-0.83

[[

-0.82

(*

-0.82

ãĢİ

-0.81

Âł

-0.81

 ****

-0.79

-"

-0.77

âĢ¦"

-0.76

 ______

-0.76

POSITIVE LOGITS

resa

1.17

zens

1.15

anmar

1.14

anamo

0.96

 RandomRedditor

0.96

Ė

0.96

ø

0.96

û

0.96

Ğ

0.96

ċ

0.96

Activations Density 0.305%

words related to expressions such as "Well," or "Do," that are commonly used at the beginning of sentences for emphasis or to introduce a new idea

conversational prompts or interjections that initiate statements or questions

No Comments

No Known Activations