INDEX

Explanations

references to "see" with a number

oai_token-act-pair · gpt-3.5-turbo

references to visual elements or instructions to "see" something

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

aced

-0.70

ascal

-0.69

EStreamFrame

-0.68

raft

-0.67

 impression

-0.65

arc

-0.65

 prog

-0.62

ament

-0.60

 educational

-0.59

 imperson

-0.58

POSITIVE LOGITS

 whence

0.92

see

0.84

geoning

0.81

agascar

0.77

¿½

0.76

ei

0.75

ethe

0.74

 supra

0.74

terday

0.74

gotten

0.74

Activations Density 0.005%

references to "see" with a number

references to visual elements or instructions to "see" something

No Comments

No Known Activations