INDEX

Explanations

phrases related to commands or instructions

oai_token-act-pair · gpt-3.5-turbo

phrases that indicate warnings or prohibitions

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 effic

-0.62

 basics

-0.60

 exemplary

-0.57

 wonderfully

-0.56

rocal

-0.56

 admirable

-0.55

 excellent

-0.55

 awesome

-0.54

 unparalleled

-0.54

 amazing

-0.54

POSITIVE LOGITS

 anymore

1.88

 unless

1.79

unless

1.66

nor

1.41

 lest

1.36

 until

1.33

until

1.33

 because

1.28

 anytime

1.26

 whatsoever

1.21

Activations Density 0.738%

phrases related to commands or instructions

phrases that indicate warnings or prohibitions

No Comments

No Known Activations