INDEX

Explanations

the phrase "question-and-answer".

oai_token-act-pair · gemini-1.5-pro Triggered by @bot

the phrase "question-and-answer", possibly with hyphens interspersed.

oai_token-act-pair · gemini-1.5-pro Triggered by @bot

phrases related to Q&A sessions or dialogues involving opinions and discussions

oai_token-act-pair · gpt-3.5-turbo

the phrase "question and answer".

oai_token-act-pair · gemini-1.5-flash Triggered by @bot

phrases related to "question and answer" sessions or formats.

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

connective terms and phrases that indicate interaction and dialogue

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 9-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.9.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.9.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

Universal

-0.80

 Tycoon

-0.77

 Deluxe

-0.77

isoft

-0.75

culus

-0.74

bourne

-0.71

gypt

-0.69

cale

-0.68

 Warehouse

-0.67

 Reserve

-0.67

POSITIVE LOGITS

 prayers

1.28

 whispers

1.11

 insults

1.10

 chants

1.10

 gestures

1.05

 messages

1.04

 applause

1.04

 jokes

1.02

 responses