INDEX

Explanations

condition-related terms such as "conditional" or "unconditional"

oai_token-act-pair · gpt-3.5-turbo

terms associated with conditions, particularly those involving conditionality in various contexts

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

 wedge

-0.79

 Stealth

-0.69

 Trainer

-0.69

 flank

-0.66

 Skies

-0.65

 Moff

-0.64

 Heritage

-0.63

 Slack

-0.61

 Sage

-0.60

 Kush

-0.60

POSITIVE LOGITS

cond

4.21

condition

1.99

Cond

1.85

 uncond

1.66

 conditional

1.58

conn

1.29

 unconditional

1.27

conduct

1.27

crit

1.23

perm

1.22

Activations Density 0.033%

condition-related terms such as "conditional" or "unconditional"

terms associated with conditions, particularly those involving conditionality in various contexts

No Comments

No Known Activations