INDEX

Explanations

usernames and social media activity

oai_token-act-pair · gpt-3.5-turbo Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs24576-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_24576

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ierrez

-0.75

 divest

-0.68

 Ministers

-0.68

enza

-0.67

 deregulation

-0.66

waukee

-0.66

 vaccines

-0.63

è¡

-0.63

 bottled

-0.61

ospital

-0.61

POSITIVE LOGITS

Redditor

1.23

username

1.13

 Reddit

1.06

 commenter

1.02

 Gamer

0.98

 reddit

0.98

 username

0.98

Reddit

0.98

 redd

0.95

 Redd

0.94

Activations Density 0.180%

usernames and social media activity

No Comments

No Known Activations