INDEX

Explanations

words related to physical assault or violence

oai_token-act-pair · gpt-3.5-turbo

references to the word "mug" 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

BCE

-0.77

Phase

-0.76

RC

-0.70

phas

-0.69

COM

-0.68

EA

-0.68

CE

-0.68

urance

-0.67

loop

-0.67

×ľ

-0.66

POSITIVE LOGITS

mug

4.13

Mug

2.26

jug

1.10

fug

1.10

 robber

1.05

 robbed

0.99

 thug

0.98

 bust

0.91

wig

0.90

 burg

0.90

Activations Density 0.017%

words related to physical assault or violence

references to the word "mug" in various contexts

No Comments

No Known Activations