© 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

QWEN3-32B · 32-RESID-BATCHTOPK-65K · 18319 ｜ Neuronpedia

Home
Qwen3-32B
32-RESID-BATCHTOPK-65K
18319

INDEX

Explanations

weak/weakness

np_acts-logits-general · gemini-2.5-flash-lite

New Auto-Interp

Top Features by Cosine Similarity

Configuration

adamkarvonen/qwen3-32b-saes/saes_Qwen_Qwen3-32B_batch_top_k/resid_post_layer_32

Prompts (Dashboard)

16,384 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

No Configuration Found

Embeds

Show PlotsShow ExplanationShow ActivationsShow Test FieldShow SteerShow Link

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

èĻ±

-0.10

wich

-0.09

wick

-0.09

tal

-0.09

iage

-0.09

avar

-0.08

å¤§åŁİå¸Ĥ

-0.08

unami

-0.08

alon

-0.08

acter

-0.08

POSITIVE LOGITS

ly

0.12

poke

0.12

Weak

0.12

æĹłåĬĽ

0.11

 máº½

0.11

å¼±

0.11

 Weak

0.11

(er

0.10

auce

0.10

ens

0.10

Activations Density 0.027%

No Known Activations