INDEX

Explanations

Sequences of characters, predominantly names, appearing multiple times across different examples. These sequences often function as proper nouns. The orthography and character lengths are consistent, allowing for recognition of patterns across examples.

eleuther_acts_top20 · gemini-1.5-flash Triggered by @johnny

references to South Asian cultural or religious concepts, particularly those related to Hinduism

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

New Auto-Interp

Configuration

neuronpedia/gpt2-small__res_slefr-ajt/2-res_slefr-ajt

Prompts (Dashboard)

12,288 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

46,080

Data Type

torch.float32

Hook Point

blocks.2.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

apollo-research/Skylion007-openwebtext-tokenizer-gpt2

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 Tripoli

-0.79

argon

-0.78

 Baghd

-0.73

 Codes

-0.71

é¾

-0.70

abulary

-0.69

Thumbnail

-0.68

Roz

-0.67

ging

-0.67

 recapt

-0.67

POSITIVE LOGITS

WARD

0.91

 Pradesh

0.82

ike

0.81

emen

0.81

opter

0.78

ewater

0.77

etheus

0.77

itage

0.74

robe

0.74

oyd

0.73

Activations Density 0.226%

Sequences of characters, predominantly names, appearing multiple times across different examples. These sequences often function as proper nouns. The orthography and character lengths are consistent, allowing for recognition of patterns across examples.

references to South Asian cultural or religious concepts, particularly those related to Hinduism

No Comments

No Known Activations