INDEX

Explanations

phrases related to incidents, events, or news reports, potentially with a negative connotation

oai_token-act-pair · gpt-3.5-turbo

sentences that report on events or situations

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

 hust

-0.78

 tremend

-0.73

 reven

-0.73

 ascend

-0.67

 diseng

-0.67

 actu

-0.66

 addon

-0.65

 heav

-0.64

 wrench

-0.63

 whist

-0.63

POSITIVE LOGITS

 Anyway

1.17

 Lastly

1.09

<|endoftext|>

1.03

 Afterwards

1.00

 Finally

0.96

 Nonetheless

0.94

 Eventually

0.94

 However

0.93

 Nevertheless

0.93

 Additionally

0.91

Activations Density 0.668%

phrases related to incidents, events, or news reports, potentially with a negative connotation

sentences that report on events or situations

No Comments

No Known Activations