INDEX

Explanations

quotes attributed to people from news reports or interviews

oai_token-act-pair · gpt-3.5-turbo

phrases or clauses indicating the source of information

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

ptin

-0.81

ãĥ¯ãĥ³

-0.70

ãĤ¼ãĤ¦ãĤ¹

-0.70

ãĤ©

-0.68

peat

-0.67

mol

-0.66

acters

-0.63

osaurs

-0.63

otin

-0.62

à¨

-0.62

POSITIVE LOGITS

 MSNBC

1.26

 Twitter

1.19

 CNBC

1.17

 behalf

1.13

CNN

1.11

 Wednesday

1.04

 Tuesday

1.04

 Monday

1.03

 Thursday

1.03

NBC

1.01

Activations Density 0.091%

quotes attributed to people from news reports or interviews

phrases or clauses indicating the source of information

No Comments

No Known Activations