INDEX

Explanations

Roman numerals, specifically 'IV' with a slightly higher activation for 'IV' at 10 compared to 9

oai_token-act-pair · gpt-3.5-turbo

references to a specific term or abbreviation, particularly related to 'IV' and associated phrases

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 goose

-0.78

 Glen

-0.72

 pine

-0.70

head

-0.63

 Hayward

-0.63

 targets

-0.62

 target

-0.61

 head

-0.61

 Chancellor

-0.61

 coffee

-0.61

POSITIVE LOGITS

IV

4.25

iv

2.22

IVES

2.08

IVE

1.95

III

1.94

VII

1.72

VI

1.69

IVER

1.67

IV

1.66

II

1.63

Activations Density 0.005%

Roman numerals, specifically 'IV' with a slightly higher activation for 'IV' at 10 compared to 9

references to a specific term or abbreviation, particularly related to 'IV' and associated phrases

No Comments

No Known Activations