INDEX

Explanations

words related to written correspondence and typography, particularly focusing on the word "letter."

oai_token-act-pair · gpt-3.5-turbo

instances of the word "letter"

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

 tranqu

-0.72

 immortality

-0.66

 prom

-0.64

 Park

-0.62

 modesty

-0.61

å¹

-0.61

 limits

-0.60

 poles

-0.58

è¿

-0.58

 Crescent

-0.58

POSITIVE LOGITS

letter

4.43

letters

2.10

LET

1.15

Letter

1.13

adder

1.06

crew

0.96

hops

0.94

footed

0.91

digit

0.91

words

0.91

Activations Density 0.021%

words related to written correspondence and typography, particularly focusing on the word "letter."

instances of the word "letter"

No Comments

No Known Activations