INDEX

Explanations

horizontal terms or concepts

oai_token-act-pair · gpt-3.5-turbo

references to horizontal orientation or alignment

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

son

-0.81

 Advisor

-0.75

rament

-0.71

Imm

-0.68

 Addiction

-0.68

ulet

-0.67

Sing

-0.67

VPN

-0.67

NSA

-0.65

hetamine

-0.65

POSITIVE LOGITS

izontal

1.07

 horizontal

0.99

cipl

0.97

¶ħ

0.82

ciples

0.78

terness

0.78

 horizontally

0.77

sbm

0.74

IPM

0.74

 striped

0.71

Activations Density 0.027%

horizontal terms or concepts

references to horizontal orientation or alignment

No Comments

No Known Activations