INDEX

Explanations

result, training, pair HT, steps, position, verification, since

This neuron fires strongly on the very first content token of each document snippet, effectively marking the start of a new segment.

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

بيق

0.90

均

0.82

0.80

<0x80>

0.79

ade

0.77

<0xE3>

0.74

asure

0.73

今日

0.73

本日

0.72

POSITIVE LOGITS

İN

0.94

IENTE

0.94

PERTIES

0.93

ITATION

0.93

ARDS

0.92

 inadequ

0.91

 LANC

0.89

AMENTE

0.89

ATIVA

0.88

CITATION

0.88

Activations Density 0.000%