INDEX

Explanations

terms related to emerging or standout performance or items

oai_token-act-pair · gpt-3.5-turbo

references to breakout performances or successes in various contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

 Dull

-0.79

rex

-0.77

 Yelp

-0.72

Pug

-0.70

 Buddh

-0.69

âĶĢ

-0.68

 Fine

-0.65

Rex

-0.65

 Misc

-0.64

opa

-0.64

POSITIVE LOGITS

 breakout

2.59

 ensemble

1.50

 crossover

1.39

 runaway

1.38

 sleeper

1.37

ilee

1.03

 surprises

1.03

ramer

0.98

 endurance

0.98

 smash

0.98

Activations Density 0.064%

terms related to emerging or standout performance or items

references to breakout performances or successes in various contexts

No Comments

No Known Activations