INDEX

Explanations

patterns specific to text segmentation markers, particularly the character sequence "Ċ"

oai_token-act-pair · gpt-3.5-turbo

instances of potentially harmful or controversial actions and situations

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

hement

-0.74

ŃĶ

-0.72

ouble

-0.71

ulkan

-0.70

ername

-0.64

ĸļ

-0.62

wart

-0.62

Ļ½

-0.62

Ĥ¬

-0.61

yssey

-0.61

POSITIVE LOGITS

è¦ļéĨĴ

0.71

egg

0.71

evidence

0.70

SOURCE

0.70

Reason

0.70

eating

0.68

Asset

0.68

Posted

0.68

credit

0.67

Saharan

0.66

Activations Density 0.489%

patterns specific to text segmentation markers, particularly the character sequence "Ċ"

instances of potentially harmful or controversial actions and situations

No Comments

No Known Activations