INDEX

Explanations

hyphenated words or phrases which emphasize a negative or contrasting element

oai_token-act-pair · gpt-3.5-turbo

negative phrases or sentiments

oai_token-act-pair · gpt-4o-mini

Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.8.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Negative Logits

 Savings

-0.69

 clause

-0.68

 Tsukuyomi

-0.68

 Improvement

-0.67

 expansion

-0.66

 portion

-0.65

 logo

-0.63

 deduction

-0.63

ulhu

-0.63

 lineup

-0.62

POSITIVE LOGITS

handedly

1.15

enough

1.06

tested

1.04

heartedly

1.04

eligible

1.03

wise

1.02

assed

0.98

compatible

0.96

shed

0.92

eyed

0.92

Activations Density 0.161%

hyphenated words or phrases which emphasize a negative or contrasting element

negative phrases or sentiments

No Known Activations