INDEX

Explanations

names, particularly the name "Dre"

oai_token-act-pair · gpt-3.5-turbo

references to specific individuals, particularly prominent names in the context of a discussion or narrative

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

names or titles starting with "Dre"

oai_token-act-pair · gpt-4-turbo

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 2-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.2.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.2.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

Department

-0.73

 Prevent

-0.69

 Absent

-0.67

 Georgian

-0.66

 Uganda

-0.66

 Hogwarts

-0.66

 Monetary

-0.65

 Administrative

-0.65

 Labour

-0.61

 Civil

-0.61

POSITIVE LOGITS

Dre

1.09

essen

1.00

acher

0.82

isson

0.80

wcs

0.77

anche

0.77

holes

0.77

aline

0.75

zi

0.75

imore

0.74

Activations Density 0.005%

names, particularly the name "Dre"

references to specific individuals, particularly prominent names in the context of a discussion or narrative

names or titles starting with "Dre"

No Comments

No Known Activations