INDEX

Explanations

instances where the audience is encouraged to engage or interact with content

oai_token-act-pair · gpt-3.5-turbo

expressions encouraging openness and sharing of thoughts or experiences

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

 burial

-0.67

 chain

-0.66

 restoration

-0.65

 satellite

-0.65

 satellites

-0.62

endi

-0.62

 improperly

-0.62

 chains

-0.61

osta

-0.61

 proposition

-0.61

POSITIVE LOGITS

 Feel

3.59

Feel

2.81

feel

1.92

 Feeling

1.66

 Views

1.25

 Taste

1.16

 Want

1.11

 Happiness

1.04

 Anger

1.03

 Fear

1.03

Activations Density 0.027%

instances where the audience is encouraged to engage or interact with content

expressions encouraging openness and sharing of thoughts or experiences

No Comments

No Known Activations