INDEX

Explanations

phrases related to costs or pricing

oai_token-act-pair · gpt-3.5-turbo

the term "high" in various contexts, often emphasizing elevated levels or quality

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

henko

-0.76

NER

-0.75

wright

-0.73

enment

-0.71

isans

-0.69

_.

-0.69

olan

-0.68

ainer

-0.68

ocide

-0.67

wen

-0.66

POSITIVE LOGITS

 high

3.15

high

2.33

 High

2.27

low

2.20

High

2.15

 HIGH

2.14

 higher

1.86

 highest

1.71

 highs

1.57

higher

1.52

Activations Density 0.038%

phrases related to costs or pricing

the term "high" in various contexts, often emphasizing elevated levels or quality

No Comments

No Known Activations