quiz bu lëkkale ak tegtal · Diggante Niveau
Laajteb Grupp
Xoolaat ni grupu-laajte bàyyi xel di séddoo boppu caabi / valeur, wàññi njëgu cache generation-time, te mën nañu ko am ci checkpoints yi fi nekk.
Laaj 1 de 8
Ci Grouped-Query Attention, lan lañu bokk ci mbooloo mu boppu laaj yi?
Xoolaatal laaj bu nekk8 laaj
Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.
- Ci Grouped-Query Attention, lan lañu bokk ci mbooloo mu boppu laaj yi?
- Caabi ak boppu valeur
- Projection biy génn kese
- duggal yi
- Tolluwaayu jàng
- GQA mooy gëna mëna wax ni diggu suuf la ci diggante ban ñaari extrême?
- Konvolusioŋ ak bàyyi xel
- Fàttaliku bu mat sëkk ci bopp yu bari ak bàyyi xel ci laaj yu bari
- FP32 ak FP8 njub
- Modèle encodeur ak decodeur kese
- Ban wàll ci inference la GQA gëna xéewale?
- Tokenisation bu laaj bi
- calcul perte de formation
- Memoire cache KV ak bandwidth bi ci diiru defar
- Softmax bi mujj ci kaw vokabileer bi
- Sudee benn model amna 32 boppu laaj ak 8 grupu KV, cache KV bi dafay wàññeeku ci lu tollu ci ban facteur?
- Lu tollu ci 32x
- Lu tollu ci 2x
- Amul benn wàññi
- Lu tollu ci 4x
- 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
- naka lañuy faral di soppi xeetu bopp yu bari yi nekk GQA?
- Soo efaase bépp couche de attention
- Ci ab jéego bu gàtt 'jëfandikoo' joge ci barabu saytu bi fi nekk
- Soo soppi ci beneen tokenizer
- Mënu leen soppi dara
- 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
- How much of the original pretraining compute did the GQA paper report using for its uptraining recipe?
- Lu tollu ci 50%
- About 100%
- No additional training compute
- Lu tollu ci 5%