INDEX

Explanations

the word "pro" or its derivatives

oai_token-act-pair · gpt-3.5-turbo

mentions of "pro" as a prefix or in relation to professional contexts

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

aji

-0.86

heit

-0.82

 Norn

-0.77

 Shades

-0.76

haven

-0.75

 Seas

-0.71

aways

-0.71

 Explorer

-0.70

imensional

-0.70

Sund

-0.69

POSITIVE LOGITS

pro

3.48

Pro

1.86

 pros

1.66

Pro

1.57

pro

1.57

 anti

1.44

PRO

1.43

 professional

1.33

 Pros

1.13

 prog

1.10

Activations Density 0.011%

the word "pro" or its derivatives

mentions of "pro" as a prefix or in relation to professional contexts

No Comments

No Known Activations