Ikibazo gihuza ibibazo · Biragoye Urwego

Ikibazo cya Mamba na Leta Yatoranijwe

Ongera usuzume Mamba ibyinjijwe-bishingiye kuri SSM ibipimo, ingano-yubunini bwa leta, umurongo-wigihe ukurikirana, hamwe nibikoresho-bimenyekanisha parallel scan.

Inzira zijyanye no kuyoboraUmwanya wa Leta ya Mamba
Ikibazo 1 Bya 8

Ni izihe nyungu nyamukuru zo kubara za Mamba kurenza Transformer isanzwe?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Ni izihe nyungu nyamukuru zo kubara za Mamba kurenza Transformer isanzwe?
    • Ipima umurongo hamwe n'uburebure bukurikiranye aho kuba kane
    • Ntabwo ikeneye amakuru yamahugurwa
    • Ikora kumashusho gusa
    • Ikuraho ibikenewe byose leta ihishe
  2. Ijambo 'guhitamo' risobanura iki muri Mamba?
    • Ihitamo GPU yo gukora
    • Ibipimo bya SSM biterwa nibyinjijwe, bityo irashobora guhitamo ibyo kwibuka cyangwa kwibagirwa
    • Ihitamo amazina ninshinga gusa
    • Ihitamo igice cyihariye cyibimenyetso kugirango wirengagize
  3. Nigute icyitegererezo cya leta cyerekana amateka yurukurikirane?
    • Kubika ibimenyetso byose byabanjirije
    • Mugushushanya ikarita yibitekerezo kuri babiri
    • Mugukusanya amateka muburyo bugaragara-bwihishe leta ivugururwa intambwe ku yindi
    • Ntabwo ibika amateka
  4. Kuki Mamba adashobora gukoresha shortcut yihuta ya SSMs yakoresheje?
    • Kuberako imyanzuro itemewe kuri GPUs
    • Kuberako idafite ibipimo
    • Kuberako itunganya ibimenyetso muburyo butunguranye
    • Kuberako gukora ibipimo byinjira-biterwa nigihe-invariance imyanzuro isaba
  5. Nubuhe buryo Mamba akoresha kugirango agume vuba nubwo yatakaje shortcut?
    • Icyuma-kimenya parallel scan kibitswe mububiko bwihuse bwa GPU
    • Kwiruka rwose kuri CPU
    • Gufata ibimenyetso byose
    • Gusimbuka byinshi murukurikirane
  6. Ninde watangije Mamba kandi hafi ryari?
    • OpenAI muri 2018
    • Albert Gu na Tri Dao mu mpera za 2023
    • Google muri 2015
    • Ntamuntu, ni hypothetical
  7. Which Mamba state-space parameters are made functions of the current input?
    • Only the fixed transition matrix A
    • The output vocabulary and tokenizer
    • B, C, and the step size delta
    • The attention query and key heads
  8. What do the variable B, C, and delta parameters let Mamba do?
    • Build a growing key-value cache for every token
    • Route each token to an expert network only
    • Replace the state update with full all-pairs attention
    • Choose how much information to let into the state and how much to retain based on the input