© Neuronpedia 2026

Privacy & Terms Blog GitHub Slack Twitter Contact

Neuronpedia

Jacobian LensNEW

Natural Language

NEW Assistant AxisNEW Circuit TracerUPDATESteer SAE Evals Exports interp-engineNEWAPI Community Blog Privacy & Terms Contact

LLAMA3-8B-IT · 25-RES-JH · 25693 ｜ Neuronpedia

Home
Julius Han · Gated SAE for Llama3-8B-Instruct
Llama3-8B-IT
Residual Stream
25-RES-JH
25693

INDEX

Explanations

the letter 'u'

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

New Auto-Interp

Top Features by Cosine Similarity

Configuration

Juliushanhanhan/llama-3-8b-it-res/blocks.25.hook_resid_post

Features

65,536

Data Type

float32

Hook Name

blocks.25.hook_resid_post

Hook Layer

25

Architecture

gated

Context Size

1,024

Dataset

Juliushanhanhan/openwebtext-1b-llama3-tokenized-cxt-1024

Activation Function

relu

Embeds

Show PlotsShow ExplanationShow ActivationsShow Test FieldShow SteerShow Link

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ez

-0.19

ej

-0.16

eam

-0.16

echn

-0.15

egie

-0.15

ï»¿↵↵

-0.15

iw

-0.15

aÄŁÄ±

-0.14

 Sears

-0.14

-0.14

POSITIVE LOGITS

ertas

0.17

alue

0.16

ltra

0.16

ALLY

0.16

Torrent

0.15

fortunately

0.15

nder

0.15

 monde

0.15

dge

0.15

ATORS

0.15

Activations Density 0.001%

No Known Activations