INDEX

Explanations

dates mentioned in emails or news articles

oai_token-act-pair · gpt-3.5-turbo

references to dates and time-related information

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

 discriminate

-0.71

estinal

-0.70

 ãĤµãĥ¼ãĥĨãĤ£ãĥ¯ãĥ³

-0.69

rats

-0.68

 ç¥ŀ

-0.68

hens

-0.65

merce

-0.64

 steril

-0.63

profits

-0.62

POSITIVE LOGITS

 circulated

0.83

 briefing

0.81

 leaked

0.81

dated

0.81

 taped

0.81

 phone

0.79

 HuffPost

0.79

 dated

0.78

 outlining

0.77

 cellphone

0.77

Activations Density 0.708%

dates mentioned in emails or news articles

references to dates and time-related information

No Comments

No Known Activations