INDEX

Explanations

terms related to dataset evaluation and model validation

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

training, validation, test sets

np_acts-logits-general · gemini-2.5-flash-lite

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k

Configuration

google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 trabajado

-0.48

 Bewußt

-0.43

 Picchu

-0.42

 cumplido

-0.42

 economía

-0.41

 pasión

-0.41

 combatir

-0.39

хьтан

-0.39

Extinguishing

-0.39

 lutte

-0.38

POSITIVE LOGITS

 ModelExpression

0.60

EndContext

0.55

migrationBuilder

0.49

 CART

0.49

elter

0.47

 conting

0.46

GOTREF

0.45

RegressionTest

0.45

quently

0.45

 getById

0.44

Activations Density 0.099%

terms related to dataset evaluation and model validation

training, validation, test sets

No Comments

No Known Activations