INDEX

Explanations

facts, numbers, statistics, and quoted statements in the context of reports, analysis, and documentation

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

•"make sure"

"suggest/note/report"

•Make sure feature

•Canonical Figure 13

No Comments

Negative Logits

gur

-0.72

 Pastebin

-0.70

ãĥĩ

-0.68

xtap

-0.65

ãĥ¬

-0.65

ciating

-0.63

ä½ľ

-0.63

quer

-0.62

osc

-0.62

adesh

-0.62

POSITIVE LOGITS

 that

1.00

 discrepancies

0.75

 inconsistencies

0.74

how

0.74

 anecd

0.72

 there

0.72

that

0.72

rists

0.71

 similarities

0.70

 they

0.69

Activations Density 3.704%

facts, numbers, statistics, and quoted statements in the context of reports, analysis, and documentation

No Comments

No Known Activations