INDEX

Explanations

start of turn indicators

This neuron detects numeric sequences (e.g. hex strings, line numbers, or file-identifier numbers) typical of error/log messages or code paths.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 אטאטורק

1.70

 hátsó

1.57

 függvény

1.56

 dizendo

1.55

 chiesto

1.55

 estavam

1.55

 Стаўкі

1.55

 cambiamenti

1.55

 අඩු

1.55

 bangunan

1.54

POSITIVE LOGITS

在

1.83

由

1.73

以

1.72

用

1.69

其

1.68

即

1.67

若

1.67

有

1.66

本

1.66

可

1.65

Activations Density 0.024%