INDEX

Explanations

strong emotional or impactful terms related to various contexts

oai_token-act-pair · gpt-3.5-turbo

references to statistical data and its implications

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 9-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.9.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.9.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

NetMessage

-0.82

entimes

-0.75

ãĤ¼ãĤ¦ãĤ¹

-0.75

Ă

-0.74

ãĤ¦ãĤ¹

-0.74

ġ

-0.73

 RandomRedditor

-0.73

ü

-0.73

÷

-0.73

Ě

-0.73

POSITIVE LOGITS

 afterwards

0.68

ITV

0.63

0.61

 Anyway

0.61

 Fifa

0.56

 Spielberg

0.56

 though

0.56

 because

0.55

 later

0.55

 Simon

0.55

Activations Density 1.274%

strong emotional or impactful terms related to various contexts

references to statistical data and its implications

No Comments

No Known Activations