INDEX

Explanations

person names and titles, potentially related to legal and criminal activities, or organized crime

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_fs1536-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_1536

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

1,536

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

mith

-0.69

 oxide

-0.68

opath

-0.65

ally

-0.64

 merit

-0.62

 terms

-0.60

ãĥ¼ãĥĨãĤ£

-0.59

alion

-0.58

 enriched

-0.58

 attest

-0.58

POSITIVE LOGITS

wait

1.04

etc

1.03

BUT

1.02

yet

0.95

until

0.94

why

0.94

 Appears

0.92

there

0.91

unless

0.91

they

0.90

Activations Density 0.783%

person names and titles, potentially related to legal and criminal activities, or organized crime

No Comments

No Known Activations