Grouped-Query Attention Quiz

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

1. MuGrouped-Query Attention, chii chinogoverwa pakati peboka remisoro yemibvunzo?
   • Misoro yakakosha uye yakakosha
   • The output projection chete
   • Iyo yekuisa embeddings
   • Chiyero chekudzidza
2. GQA inotsanangurwa zvakanyanya senzvimbo yepakati pakati pezvipi zviviri zvakanyanyisa?
   • Convolution uye kutarisa
   • Yakazara-yakawanda-yemusoro kutarisisa uye yakawanda-mubvunzo kutarisisa
   • FP32 uye FP8 chaiyo
   • Encoder-chete uye decoder-chete modhi
3. Ndeipi chikamu chekufungidzira icho GQA inonyanya kuita yakachipa?
   • Tokenization yekukurumidza
   • Iko kurasika kwekudzidzira kuverenga
   • Iyo KV cache memory uye bandwidth yayo panguva yechizvarwa
   • Iyo yekupedzisira softmax pamusoro pemazwi
4. Kana modhi iine misoro yemibvunzo makumi matatu nemaviri uye mapoka masere eKV, cache yeKV inoderedzwa neicho chinhu?
   • Kunenge 32x
   • Zvinenge 2x
   • Hapana kuderedza
   • Zvinenge 4x
5. What quality/speed tradeoff did the GQA paper report in its experiments?
   • Quality close to MHA with inference speed closer to MQA
   • Lower quality and slower inference than both
   • Exactly the same latency and quality for every model
   • A result only on image classification
6. Mamodheru aripo ane misoro mizhinji anowanzo shandurwa kuita GQA?
   • Nekubvisa ese ekutarisa maseru
   • Nekapfupi 'uptraining' nhanho kubva pane iripo yekutarisa
   • Nekuchinja kune imwe tokenizer
   • Havagoni kutendeuka zvachose
7. Which task families did the GQA paper evaluate?
   • Image segmentation and object detection
   • Speech synthesis and speaker verification
   • Language modeling and machine translation
   • Molecular generation and toxicity prediction
8. How much of the original pretraining compute did the GQA paper report using for its uptraining recipe?
   • Zvinenge 50%
   • About 100%
   • No additional training compute
   • Zvingangoita 5%