INDEX

Explanations

phrases related to news events, particularly involving incidents and victims

oai_token-act-pair · gpt-3.5-turbo

instances of the verb "was" indicating occurrences or states of being

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

 entails

-0.76

 ç¥ŀ

-0.74

poll

-0.73

inav

-0.71

 Critics

-0.67

 univers

-0.66

 Highlights

-0.66

 Preferences

-0.66

 Which

-0.65

 offsets

-0.65

POSITIVE LOGITS

 arra

0.98

 hospitalized

0.93

 supposed

0.91

 able

0.91

 reportedly

0.90

 unable

0.89

 alerted

0.88

hes

0.88

 transported

0.87

 beaten

0.86

Activations Density 0.318%

phrases related to news events, particularly involving incidents and victims

instances of the verb "was" indicating occurrences or states of being

No Comments

No Known Activations