Guide-yakabatana mibvunzo · Zvakaoma Nhanho
Attention Rollout uye Musoro Pruning Quiz
Inoedza kuti kuburitswa kunobatanidza kutariswa kwakaiswa sei nekwasara nzira, ndedzipi fungidziro yekuchekerera musoro, uye izvo zvibodzwa zvinogona uye zvisingakwanisi kutsanangura.
Mubvunzo 1 ye 8
Chinangwa chekutarisisa ndechei?
Tarisa mubvunzo wese8 mibvunzo
Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.
- Chinangwa chekutarisisa ndechei?
- Kutsvaga ruzivo rwekuyerera kubva kune ekuisa tokens kuenda kune zvinobuda muzvikamu
- Kuwedzera mamwe maseru kune Transformer
- Kuwedzera mwero wekudzidza
- Kushandura pfungwa kuita convolution
- Sei mepu yekutarisisa yechikamu chimwe isina kukwana kutsanangura?
- Attention haimboverengerwa
- Ruzivo rwunosanganiswa muzvikamu zvakawanda uye misoro, saka mepu imwe inopotsa nzira yakazara
- Mepu dzinogara dzakafanana
- Transformers vane musoro mumwe chete
- Chii chinopa kutarisisa kunowedzera kune yega yega matrix yekuzvidavirira kune yakasara yekubatanidza?
- A bias ye zero
- Izwi reruzha risina kurongeka
- Identity matrix chikamu chisati chajairika
- Imwe softmax
- Ko tsvagiridzo pamusoro pekutarisa-mumusoro-zhinji yakaratidzei nezve kuderedzwa kwemusoro?
- Musoro wese wakakosha
- Misoro haigoni kubviswa
- Kuchekerera nguva dzose kunowedzera kaviri kururama
- Misoro mizhinji inogona kudimburwa nekurasikirwa kushoma chaiko
- Kukosha kwemusoro kunowanzofungidzirwa sei pakuchekerera?
- Via gradient-based sensitivity yekurasikirwa kune mask yemusoro
- Nechinzvimbo chearufabheti
- Nekuverenga ma parameter ayo chete
- Nenguva yezuva
- Ndeapi mabasa akasiyana anoita kutarisa uye kuchekerera musoro rimwe nerimwe?
- Iwo maitiro akafanana
- Kuburitsa kunobatsira kududzira uku kuchekerera kunoshandisa redundancy kumanikidza modhi
- Ose ari maviri anowedzera ma parameter
- Kana kubata pfungwa
- In an attention-rollout calculation, why are residual paths included with layer attention?
- To make every head have the same weight
- To remove the value vectors from the Transformer
- To approximate information that can pass along the identity path when layer transitions are composed
- To prevent the attention matrix from being normalized
- How should an attention-rollout score be interpreted?
- As proof that a token caused the output
- As the exact value vector used at every layer
- As a measure of how many heads the model contains
- As a post-hoc approximation of relative token influence, not a causal explanation