INDEX

Explanations

phrases related to suggestions, actions or instructions

oai_token-act-pair · gpt-3.5-turbo

phrases indicating authority, desire, or calls to action

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

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

 Sponge

-0.71

 Survivors

-0.70

 sidx

-0.65

 Actual

-0.65

 unin

-0.61

 elapsed

-0.60

atem

-0.59

 apologised

-0.59

 implied

-0.57

 totaled

-0.57

POSITIVE LOGITS

haven

0.74

must

0.70

 enlist

0.68

 hesitate

0.68

 foss

0.66

NetMessage

0.66

 consult

0.65

Ń·

0.64

hower

0.64

went

0.64

Activations Density 0.319%

phrases related to suggestions, actions or instructions

phrases indicating authority, desire, or calls to action

No Comments

No Known Activations