INDEX

Explanations

the word segment "unw" followed by a high-value activation word component

oai_token-act-pair · gpt-3.5-turbo

references to unwritten rules or social norms

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

phrine

-0.94

ãĤ¼

-0.84

hyde

-0.81

uyomi

-0.81

senal

-0.79

pmwiki

-0.77

 Defenders

-0.73

anwhile

-0.72

 hemor

-0.72

å§«

-0.72

POSITIVE LOGITS

arranted

1.05

ield

1.04

ritten

1.04

ashed

0.99

irth

0.99

inding

0.96

ashington

0.94

atcher

0.93

itt

0.93

avering

0.92

Activations Density 0.009%

the word segment "unw" followed by a high-value activation word component

references to unwritten rules or social norms

No Comments

No Known Activations