INDEX

Explanations

fragments related to selected news or opinions

oai_token-act-pair · gpt-3.5-turbo

punctuation marks or specific phrases that signal discussions of titles or lists

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 acquaintance

-0.77

 applicant

-0.72

 inclined

-0.72

 congress

-0.71

 inhibitor

-0.70

 regist

-0.69

 allowance

-0.68

 bond

-0.68

 accumulated

-0.68

 retri

-0.67

POSITIVE LOGITS

 Including

1.51

Pt

1.40

 Stupid

1.31

 Part

1.29

 Volume

1.21

 Says

1.17

Too

1.16

 Lessons

1.16

 Featuring

1.16

 Conclusion

1.15

Activations Density 0.121%

fragments related to selected news or opinions

punctuation marks or specific phrases that signal discussions of titles or lists

No Comments

No Known Activations