INDEX

Explanations

phrases indicating reader engagement and opinions about content

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-9B @ 34-gemmascope-res-16k

Configuration

google/gemma-scope-9b-pt-res/layer_34/width_16k/average_l0_114

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

16,384

Data Type

float32

Hook Name

blocks.34.hook_resid_post

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 absolutely

-1.20

 unbelievably

-1.19

 incredibly

-1.16

 extrêmement

-1.13

 absolutamente

-1.10

absolutely

-1.09

 extremely

-1.09

 абсолютно

-1.06

 utterly

-1.05

extremely

-1.04

POSITIVE LOGITS

 somewhat

0.99

somewhat

0.92

 Somewhat

0.85

 trochu

0.84

 biraz

0.82

 trochę

0.81

 nieco

0.79

Somewhat

0.78

 perhaps

0.76

 qualche

0.75

Activations Density 3.485%

phrases indicating reader engagement and opinions about content

No Comments

No Known Activations