INDEX

Explanations

terms related to leaks and leaking information

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

New Auto-Interp

Configuration

google/gemma-scope-9b-pt-res/layer_20/width_131k/average_l0_114

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

131,072

Data Type

float32

Hook Name

blocks.20.hook_resid_post

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 CommandType

-0.55

 numberWith

-0.52

ptonshire

-0.52

 functools

-0.49

opsida

-0.49

CommandType

-0.48

Ordinal

-0.44

AtPosition

-0.44

 Afin

-0.44

 Cardinal

-0.43

POSITIVE LOGITS

 leak

1.16

 Leak

1.13

leak

1.11

Leak

1.10

 leaks

1.04

 Leaks

0.96

 leakage

0.94

 Leakage

0.93

 leaking

0.91

 leaked

0.90

Activations Density 0.008%

terms related to leaks and leaking information

No Comments

No Known Activations