INDEX

Explanations

a series of false positive activations in multiple languages, likely due to noise or unrelated text

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

identifiers or specific symbols from different languages or scripts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 2-att-kk

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L2_Hcat_z_lr1.20e-03_l11.00e+00_ds24576_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v4.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

float32

Hook Name

blocks.2.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.06

1:0.02

2:0.15

3:0.07

4:0.09

5:0.10

6:0.05

7:0.06

8:0.13

9:0.09

10:0.08

11:0.03

Negative Logits

nox

-1.27

uma

-1.26

...]

-1.25

 Janeiro

-1.21

asonic

-1.20

ength

-1.18

onductor

-1.17

Thumbnail

-1.15

terness

-1.14

rier

-1.14

POSITIVE LOGITS

blank

1.08

イト

1.07

DonaldTrump

1.07

 Targ

1.04

 pedest

1.02

stock

1.01

ibilities

0.99

 Artists

0.99

 Government

0.99

 royalties

0.99

Activations Density 0.002%

a series of false positive activations in multiple languages, likely due to noise or unrelated text

identifiers or specific symbols from different languages or scripts

No Comments

No Known Activations