INDEX

Explanations

questions or requests for information

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

ufact

-0.76

Ĥ¬

-0.75

 Scouting

-0.65

lim

-0.63

Lago

-0.61

rites

-0.61

EStreamFrame

-0.59

ofi

-0.59

luaj

-0.59

relative

-0.58

POSITIVE LOGITS

 questions

1.37

naires

1.24

 rhet

1.20

 answered

1.17

 probing

1.15

 Questions

1.13

 unanswered

1.07

 question

1.06

naire

1.06

answered

1.03

Activations Density 3.649%

questions or requests for information

No Comments

No Known Activations