INDEX

Explanations

question marks followed by line breaks in interview-style text.

oai_token-act-pair · claude-3-5-sonnet-20240620 Triggered by @danielepoterti

patterns related to the structure of a conversation or dialogue, specifically related to turn-taking and dialogue attribution

oai_token-act-pair · gpt-3.5-turbo

questions that prompt discussion or opinions

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.8.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

 casc

-0.71

 downstream

-0.71

 grav

-0.70

 existential

-0.68

 galvan

-0.66

 vicious

-0.66

 probing

-0.65

 Watergate

-0.65

 multiplied

-0.64

 cartel

-0.64

POSITIVE LOGITS

Answer

1.11

AH

1.03

gow

1.00

OA

0.97

JM

0.96

JB

0.93

SG

0.92

Yeah

0.92

VB

0.90

RM

0.90

Activations Density 0.109%

question marks followed by line breaks in interview-style text.

patterns related to the structure of a conversation or dialogue, specifically related to turn-taking and dialogue attribution

questions that prompt discussion or opinions

No Comments

No Known Activations