INDEX

Explanations

statistical and numerical data patterns

oai_token-act-pair · gpt-3.5-turbo

Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs768-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_768

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

768

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

IFrame

Link

Negative Logits

haven

-0.74

abs

-0.67

rovers

-0.66

XT

-0.66

 Scholars

-0.64

 Consulting

-0.63

 liabilities

-0.63

hawks

-0.62

masters

-0.61

stones

-0.60

POSITIVE LOGITS

 capita

1.11

 successive

1.09

 inch

1.08

 ounce

1.07

tains

1.05

 month

1.00

 person

0.95

 year

0.94

malink

0.94

 hour

0.92

Activations Density 2.612%

statistical and numerical data patterns

No Known Activations