INDEX

Explanations

proper nouns, likely related to news headlines or public figures

oai_token-act-pair · gpt-3.5-turbo

references to public figures and incidents involving controversy or conflict

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

 probabilities

-0.78

 Completed

-0.77

"],"

-0.76

chance

-0.73

 feasibility

-0.69

aceutical

-0.69

pletion

-0.69

 Prediction

-0.69

 Farming

-0.69

 completion

-0.68

POSITIVE LOGITS

 insulted

1.55

 offended

1.47

 lashed

1.34

 clashed

1.33

 harassed

1.31

 humiliated

1.28

 hurled

1.25

 criticised

1.24

 angered

1.24

 insults

1.23

Activations Density 0.681%

proper nouns, likely related to news headlines or public figures

references to public figures and incidents involving controversy or conflict

No Comments

No Known Activations