quiz bu lëkkale ak tegtal · Dëgër Niveau

Modèle espace d'État et Quiz Mamba

Saytu jafe-jafey toppalante espace-espace, tànneef Mamba, scanner xam-xam hardware, etaa inference buy baaxoo, ak xeetu hybrid.

Yooni njiit yu jëm ciModèles d'espace état et mamba
Laaj 1 de 8

naka la ko modelu espace etaa bi di doxee ak guddaayu toppalante bi, buñu ko méngale ak bàyyi xel ci sa bopp buñ miin?

Xoolaatal laaj bu nekk8 laaj

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

  1. naka la ko modelu espace etaa bi di doxee ak guddaayu toppalante bi, buñu ko méngale ak bàyyi xel ci sa bopp buñ miin?
    • Lineer, ak escaleer quadratique de l'attention
    • Ci anam wu kwadratik, ak eskalaasioŋ ligneer bu bàyyi xel
    • Ñoom ñaar ñépp dañuy nuru
    • Exponentiellement ngir ñoom ñaar
  2. Lan mooy jéego yu am solo yi Mamba yokk ci SSM yu njëkk ya melni S4?
    • Jëfandikoo softmax ci biir
    • Defar parametru yeesal etaa a ngi aju ci dugal bi (tanneef)
    • Di dindi nekkin bu nëbbu bi yépp
    • Weccoo ci xayma waxtu kwadratik
  3. Lan moo tax Mamba soxla ab scanner parallèle bu xam hardware bi?
    • Ndax GPU yi mënu ñu def matrix yu bari
    • Ngir wàññi limu diisaay yi
    • Ndax def ay parametre yu aju ci dugal dafay dog njuuj njaaj bu yomb bi ci convolution
    • Ndax jëfandikoowul ay paramet yuñ jàng
  4. Bu ñuy jël doggal, njariñu efficacité bu mag bi ci SSM bu melni Mamba mooy dafay jëfandikoo:
    • Memoire buy màgg ak guddaayi sequence bi yépp, melni cache KV
    • Amul benn fàttaliku
    • Memoire bu gëna bari Transformateur bu tollu ci dayo
    • Lu tollu ci mémoire bu soppeeku ci token bu nekk
  5. Ci SSM yuñ yamale (balaa tànneef), mën nañu xayma recurrence bi bu baax ndax mën nañu ko binndaat ni:
    • Konvolusioŋ bu gudd
    • Liggéeyukaay Softmax
    • Garab dogal
    • àll bu bari
  6. Which language-modeling corpus did the Mamba paper use in its standard autoregressive scaling experiments?
    • ImageNet-1K
    • Pile bi
    • LibriSpeech only
    • D4RL
  7. In the Mamba block, what operation combines the selective-SSM output with the other projected branch?
    • A nearest-neighbor lookup over audio frames
    • A cross-attention layer over image patches
    • An elementwise multiplicative interaction with a gated branch
    • A graph-degree normalization step
  8. What language-modeling result does the Mamba paper report for its 3B model?
    • It requires a Transformer cache twice as large
    • It replaces the training corpus with audio only
    • It can be trained without any data
    • It matches Transformers twice its size in pretraining and downstream evaluation