INDEX

Explanations

This attention head attends to the first content token after the <|endoftext|> token from tokens later in the sequence.

oai_attention-head · claude-3-7-sonnet-20250219 Triggered by @trainjumperx

references to specific people and their actions or statuses

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 9-att-kk

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L9_Hcat_z_lr1.20e-03_l11.20e+00_ds24576_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v9.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

float32

Hook Name

blocks.9.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.08

1:0.02

2:0.07

3:0.04

4:0.03

5:0.04

6:0.24

7:0.04

8:0.06

9:0.25

10:0.02

11:0.04

Negative Logits

Wag

-4.05

 Bach

-3.82

 goats

-3.80

 cous

-3.78

bush

-3.68

 Hercules

-3.63

 crochet

-3.58

 cater

-3.52

 horns

-3.51

 Circus

-3.37

POSITIVE LOGITS

Len

10.37

Len

8.94

 Lenn

5.33

len

5.09

Lip

4.53

 Leonard

4.50

len

4.19

 Staten

4.13

 Lowe

4.11

 Lynch

3.99

Activations Density 0.001%

This attention head attends to the first content token after the <|endoftext|> token from tokens later in the sequence.

references to specific people and their actions or statuses

No Comments

No Known Activations