INDEX

Explanations

references to watching or monitoring activities

oai_token-act-pair · gpt-3.5-turbo

references to the organization "Watch."

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

 routed

-0.63

 phased

-0.62

urance

-0.62

 depreciation

-0.62

alyses

-0.61

 Sakuya

-0.61

cific

-0.61

effect

-0.61

 suffix

-0.60

meaning

-0.59

POSITIVE LOGITS

 Watch

3.97

watch

2.53

Watch

2.52

 WATCH

2.39

 watch

2.20

 watches

1.73

 Watching

1.62

WATCH

1.58

Wat

1.49

wat

1.35

Activations Density 0.013%

references to watching or monitoring activities

references to the organization "Watch."

No Comments

No Known Activations