quiz bu lëkkale ak tegtal · Dëgër Niveau
Autoencoder yu néew ngir laaj tekki
Saytu defaraat ak sparsity autoencoder bu bariwul, man-mani aktivasioŋ, yaatuwaayu yaatuwaayu, xaaj man-mani ak jàngat ci direction causal.
Laaj 1 de 8
Lan mooy jubluwaay bi gëna mag ci tàggat ab autoencoder bu néew ci ab model buy tàmbali?
Xoolaatal laaj bu nekk8 laaj
Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.
- Lan mooy jubluwaay bi gëna mag ci tàggat ab autoencoder bu néew ci ab model buy tàmbali?
- Ngir xaaj aktivaasioŋ yu jaxasoo yi ci màndarga yuñ mëna tekki, yu gëna bari ay màndarga yuñ tàqale
- ngir kompresse model bi mu gëna gaaw
- Tekki mbind ci diggante làkk yi
- Yokk njubte gi ci model bi ci benchmark yi
- How does an SAE’s sparsity penalty affect hidden-unit activations?
- It guarantees each unit represents exactly one concept
- It encourages only a small subset of the wider hidden layer to activate for an input
- It makes every hidden unit activate equally
- It removes the reconstruction objective
- Lu tollu ci ñaata man-man la Anthropic génne bi muy yokk SAEs ci Claude 3 Sonet ci 2024?
- Lu tollu ci 500
- Lu tollu ci 2000
- Lu tollu ci 34 milioŋ
- Lu tollu ci 100 milyaar
- Lan la 'Buntu wurus Claude' wane?
- Ni SAEs mën nañu leen tàggat ci nataal yi
- Ni model bi mënatul defar mbind
- Man-man yi amul benn njeexital ci génne yi
- Ni clamping benn manndarga bu kawe sabab bi dafay doxal doxalinu model bi
- Lan moo waral layer bu nëbbu ci SAE di gëna yaatu ak activation bi muy defaraat?
- Ndax konsept yi dañu leen boole ci dimension yu néew yu gëna néew li ñu soxla, kon dañu leen di jàppale ñu dindi leen
- Ngir yombal tàggat yaram
- Ndax diisaay yu gëna yaatu ñooy gëna dëggu
- Ngir wàññi limu parametre yi
- Lan mooy yamaleg autoencoder yu bari yi am ci jamono jii?
- Mënu leen tàggat ci wàccinu gradient
- Xaajale màndarga, fu benn konsept xaajaloo nekk màndarga yu bari yu wuute
- Dañuy liggéey ci reso convolutionnel kese
- Dañu dindi bépp etiketu nit
- Which two objectives are commonly combined when training a sparse autoencoder?
- A policy gradient and value loss
- A next-token loss and a contrastive loss
- Reconstruction error and a penalty on hidden activations
- A calibration error and classification accuracy
- What does a TopK sparse autoencoder do to encourage sparsity?
- Keeps the K largest input pixels and discards the decoder
- Sets all encoder weights to zero
- Trains K different tokenizers
- Keeps only the K largest hidden activations for an input