INDEX

Explanations

the word "watch" in various contexts

oai_token-act-pair · gpt-3.5-turbo

mentions of the verb "watch."

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

interstitial

-0.74

 misunderstanding

-0.72

ãĤ¨ãĥ«

-0.69

pse

-0.66

 hemorrh

-0.66

activation

-0.66

phi

-0.65

ãĥ´ãĤ¡

-0.64

xual

-0.64

ctrl

-0.63

POSITIVE LOGITS

tower

1.28

 watch

1.15

 Watching

1.09

 watches

1.06

 Watch

1.04

watch

1.01

 watching

0.94

dogs

0.91

Watch

0.88

 watched

0.88

Activations Density 0.014%

the word "watch" in various contexts

mentions of the verb "watch."

No Comments

No Known Activations