Kedis ku xidhan hagaha · Adag Heerka

Models Space State iyo Mamba Kediska

Tijaabi kakanaanta isku xigxiga-meel-goboleedka, xulashada Mamba, scan-ka warhaynta qalabka, xaaladda soo noqnoqda, iyo moodooyinka isku-dhafan.

Wadooyinka haga ee la xidhiidhaModels Space State iyo Mamba
Su'aal 1 ee 8

Sidee buu u miisaamayaa qaabka booska gobolka ee leh dhererka isku xigxiga, marka la barbar dhigo is-fiirinta caadiga ah?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Sidee buu u miisaamayaa qaabka booska gobolka ee leh dhererka isku xigxiga, marka la barbar dhigo is-fiirinta caadiga ah?
    • Si toos ah, oo ka soo horjeeda miisaanka quadratic ee dareenka
    • Afar ahaan, ka soo horjeedka miisaanka toosan ee dareenka
    • Labaduba waa isku mid
    • Si qaas ah labadaba
  2. Waa maxay hal-abuurka dhexe ee uu Mamba ku daray SSM-yadii hore ee habaysan sida S4?
    • Isticmaalka dareenka softmax gudaha
    • Samaynta halbeegyada cusboonaysiinta gobolka waxay kuxirantahay gelinta (doorashada)
    • Gebi ahaanba meesha laga saarayo gobolka qarsoon
    • U beddelashada xisaabinta-waqtiga-quadratic-ka
  3. Muxuu Mamba ugu baahday iskaanka isbarbar socda ee hardware-ka warqaba ee gaarka ah?
    • Sababtoo ah GPU-yadu ma samayn karaan isku dhufashada matrixka
    • Si loo yareeyo tirada lakabyada
    • Sababtoo ah samaynta cabbiraadaha ku-tiirsanaanta waxay jebiyaa khiyaamada is-beddelka ee fudud
    • Sababtoo ah ma isticmaaleyso cabbiro la bartay
  4. Inta lagu jiro falanqaynta, faa'iidada muhiimka ah ee waxtarka leh ee SSM sida Mamba waa inay isticmaasho:
    • Xusuusta ku koraysa dhererka taxanaha buuxa, sida kaydka KV
    • Wax xasuus ah haba yaraatee
    • Xusuus badan oo ka badan Transformer cabbir siman
    • Qiyaasta xusuusta joogtada ah ee calaamad kasta
  5. SSM-yada habaysan ee caadiga ah (kahor xulashada), soo noqoshada si hufan ayaa loo xisaabin karaa sababtoo ah dib ayaa loo qori karaa sida:
    • Kacdoon dheer
    • Hawlgalka Softmax
    • Geedka go'aanka
    • Kaynta random
  6. Which language-modeling corpus did the Mamba paper use in its standard autoregressive scaling experiments?
    • ImageNet-1K
    • Tusmada
    • 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