INDEX

Explanations

years in the 1998

oai_token-act-pair · gpt-3.5-turbo

mentions of the year 1998

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

Token

-0.87

-0.70

sab

-0.69

Ĩ

-0.68

antry

-0.68

abb

-0.67

 sounded

-0.65

qt

-0.64

Medium

-0.64

MU

-0.64

POSITIVE LOGITS

2.94

2.90

2.71

2.66

2.40

2.19

2.17

2.12

2.06

Activations Density 0.030%

years in the 1998

mentions of the year 1998

No Comments

No Known Activations