INDEX

Explanations

any word related to legal matters

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

ãĤ¨ãĥ«

-0.98

ãĥĥ

-0.74

bats

-0.74

angible

-0.73

RON

-0.72

uga

-0.71

ynt

-0.71

cash

-0.71

tmp

-0.70

BUS

-0.70

POSITIVE LOGITS

 syndrome

0.86

âĦ¢:

0.83

 Productions

0.79

 Syndrome

0.73

 equivalents

0.72

 approaches

0.72

 scenarios

0.72

 pamph

0.70

 showdown

0.70

 scenario

0.69

Activations Density 6.281%

any word related to legal matters

No Comments

No Known Activations