INDEX

Explanations

sequences of numbers separated by underscores and possibly followed by other characters

oai_token-act-pair · gpt-3.5-turbo

sequences of numerical data, possibly related to file names or codes

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.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

 compulsion

-0.76

ossibility

-0.69

 belts

-0.69

 Virtue

-0.67

 signs

-0.67

 Golem

-0.62

ortunately

-0.62

theless

-0.62

rencies

-0.61

 Leap

-0.59

POSITIVE LOGITS

_-_

1.24

1.15

_-

1.10

__

1.03

_.

1.01

0.95

0.94

jpg

0.92

"}],"

0.89

"},{"

0.89

Activations Density 0.082%

sequences of numbers separated by underscores and possibly followed by other characters

sequences of numerical data, possibly related to file names or codes

No Comments

No Known Activations