INDEX

Explanations

phrases related to things that are left or that remain

oai_token-act-pair · gpt-3.5-turbo

phrases that indicate something is left or still exists

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

arta

-0.73

 Babe

-0.68

 Topic

-0.66

icz

-0.66

jah

-0.63

think

-0.62

LI

-0.58

gradient

-0.57

forum

-0.57

dos

-0.56

POSITIVE LOGITS

 remaining

3.78

 remainder

2.27

 surviving

1.91

 leftover

1.44

 lingering

1.35

 remain

1.32

 rest

1.31

 remained

1.25

 residual

1.23

 retaining

1.23

Activations Density 0.005%

phrases related to things that are left or that remain

phrases that indicate something is left or still exists

No Comments

No Known Activations