INDEX

Explanations

content related to harmful or offensive behavior

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 10-res_post_32k-oai

Configuration

jbloom/GPT2-Small-OAI-v5-32k-resid-post-SAEs/v5_32k_layer_10.pt

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

32,768

Data Type

torch.float32

Hook Name

blocks.10.hook_resid_post

Hook Layer

Architecture

standard

Context Size

Dataset

Skylion007/openwebtext

Activation Function

topk

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

escription

-0.91

terday

-0.80

pection

-0.76

gements

-0.75

rica

-0.71

vae

-0.67

gem

-0.67

ablishment

-0.67

ciples

-0.66

orses

-0.65

POSITIVE LOGITS

 negligence

0.68

 inaction

0.64

 Electronics

0.61

 Squid

0.60

 Router

0.60

 breakdown

0.59

 intentionally

0.58

 caused

0.57

 rage

0.57

 Spray

0.57

Activations Density 0.087%

content related to harmful or offensive behavior

No Comments

No Known Activations