INDEX

Explanations

dates, specifically in the form of "Month Day"

oai_token-act-pair · gpt-3.5-turbo

occurrences of the word "January" followed by numerical dates

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

 orcs

-0.65

aho

-0.64

 glances

-0.62

ography

-0.61

ific

-0.60

ologies

-0.59

 absorbs

-0.59

lov

-0.59

 totem

-0.58

ileged

-0.57

POSITIVE LOGITS

 January

3.21

 February

2.75

January

2.66

 December

2.60

 November

2.52

 October

2.41

 July

2.40

 September

2.34

 March

2.34

 April

2.29

Activations Density 0.015%

dates, specifically in the form of "Month Day"

occurrences of the word "January" followed by numerical dates

No Comments

No Known Activations