INDEX

Explanations

words related to physical or emotional harm or mistreatment

oai_token-act-pair · gpt-3.5-turbo

references to various forms of abuse

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 completion

-0.71

 countdown

-0.67

tern

-0.63

 align

-0.63

hua

-0.63

 graduation

-0.63

Enlarge

-0.63

 Plan

-0.62

ence

-0.61

ty

-0.61

POSITIVE LOGITS

 abused

3.77

 abusing

1.91

 exploited

1.77

 abuser

1.70

 persecuted

1.61

 victimized

1.61

 harassed

1.60

 assaulted

1.60

 tortured

1.59

 abuse

1.59

Activations Density 0.017%

words related to physical or emotional harm or mistreatment

references to various forms of abuse

No Comments

No Known Activations