INDEX

Explanations

words related to acknowledging or confessing something

oai_token-act-pair · gpt-3.5-turbo

instances of the word "admit" and its variations in different contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

srf

-0.81

 impacted

-0.78

ammy

-0.78

 clipboard

-0.75

 chimpan

-0.75

 tremend

-0.75

ccording

-0.74

pleting

-0.71

 newcom

-0.71

 effected

-0.71

POSITIVE LOGITS

 admit

1.17

ibility

1.05

 admitted

0.94

 admits

0.94

 admitting

0.90

iary

0.88

essors

0.88

 Admission

0.82

galitarian

0.81

 admission

0.78

Activations Density 0.011%

words related to acknowledging or confessing something

instances of the word "admit" and its variations in different contexts

No Comments

No Known Activations