INDEX

Explanations

words related to personal preferences and interactions

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_fs3072-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_3072

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

3,072

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ilion

-0.81

arta

-0.81

aunder

-0.78

ourse

-0.69

ocument

-0.69

DragonMagazine

-0.67

vernment

-0.66

ItemImage

-0.66

emin

-0.65

WF

-0.64

POSITIVE LOGITS

 seeing

0.96

 watching

0.95

 hearing

0.84

 spicy

0.81

 dearly

0.80

ably

0.79

 surprises

0.77

 listening

0.76

lihood

0.75

 tink

0.73

Activations Density 0.933%

words related to personal preferences and interactions

No Comments

No Known Activations