INDEX

Explanations

hyphenated numerical values that seem to relate to some form of quantity or amount

oai_token-act-pair · gpt-3.5-turbo

negative expressions or contexts, particularly involving duration or quantity

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

 Moody

-0.74

 cues

-0.73

 Tuls

-0.70

WD

-0.69

 Ernst

-0.68

 sights

-0.68

 infl

-0.68

 Rhod

-0.67

 irregularities

-0.67

 tabs

-0.66

POSITIVE LOGITS

century

0.98

ccording

0.95

degree

0.95

series

0.93

sized

0.92

year

0.92

minute

0.92

dozen

0.92

orbit

0.91

factor

0.91

Activations Density 0.032%

hyphenated numerical values that seem to relate to some form of quantity or amount

negative expressions or contexts, particularly involving duration or quantity

No Comments

No Known Activations