INDEX

Explanations

phrases expressing strong personal preferences or identities

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_post_32k-oai

Configuration

jbloom/GPT2-Small-OAI-v5-32k-resid-post-SAEs/v5_32k_layer_8.pt

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

32,768

Data Type

torch.float32

Hook Name

blocks.8.hook_resid_post

Hook Layer

Architecture

standard

Context Size

Dataset

Skylion007/openwebtext

Activation Function

topk

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 menacing

-0.80

 unrecogn

-0.72

 majesty

-0.72

 overshadow

-0.71

 incrim

-0.70

 unheard

-0.70

 assassinate

-0.70

 menace

-0.69

 virtues

-0.68

 believable

-0.66

POSITIVE LOGITS

math

0.78

consumer

0.74

OCD

0.73

 betting

0.69

intend

0.68

 sucker

0.68

fan

0.66

 avid

0.66

chool

0.65

price

0.65

Activations Density 0.488%

phrases expressing strong personal preferences or identities

No Comments

No Known Activations