INDEX

Explanations

words indicating numerical or coded specifications

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

TagMode

-0.53

hysema

-0.45

новниш

-0.44

 برانيه

-0.44

hubanes

-0.43

unicaciones

-0.41

dighet

-0.40

ształ

-0.40

andle

-0.40

 Jawahar

-0.40

POSITIVE LOGITS

Autoritní

0.59

fromnode

0.44

pet

0.41

Pet

0.39

Sucesor

0.39

 betweenstory

0.38

 pets

0.38

 tray

0.37

 informée

0.37

Activations Density 0.091%

words indicating numerical or coded specifications

No Comments

No Known Activations