INDEX

Explanations

words related to emotions, opinions, and personal 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_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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

EStream

-0.78

å§«

-0.77

NESS

-0.75

 resil

-0.74

 shorth

-0.68

 Mechdragon

-0.68

 Lauder

-0.68

 ãĤµãĥ¼ãĥĨãĤ£ãĥ¯ãĥ³

-0.67

 Mellon

-0.66

-0.64

POSITIVE LOGITS

oused

1.06

agn

1.06

angs

1.06

agging

1.05

umbling

1.04

unk

1.04

umbled

1.03

ink

1.02

apped

1.02

ifts

1.02

Activations Density 2.706%

words related to emotions, opinions, and personal interactions

No Comments

No Known Activations