SmoothQuant uye Activation Quantization Quiz

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

1. Nderipi dambudziko iro SmoothQuant rinonyatso gadzirisa mune yakaderera-chaiyo inference?
   • Yakanyanya outlier kukosha mu activation channels
   • Huremu hudiki kuti humiririre
   • Gradient inonyangarika panguva yekudzokera shure
   • Chiratidzo chemazwi saizi miganho
2. SmoothQuant inoita sei kuti ma activation ave nyore kuverengera?
   • Iyo inocheka zvese activation pamusoro peyakagadziriswa chikumbaridzo
   • Inotamisa kuoma nekudzika ma activations pasi uye kuyera kukwira nechinhu chimwe chete
   • Iyo inodzokorora iyo modhi yekupedzisira-kusvika-kumagumo mu8-bit
   • Inoshandura ma activation kuita inoyangarara point chete
3. Ko notation W8A8 inorevei?
   • 8 uremu akaturikidzana uye 8 yekutarisisa misoro
   • Kuora kwehuremu hwe8 uye alpha ye8
   • 8-bit uremu uye 8-bit activation
   • 8 vashandi uye 8 accelerators
4. Sei kuyera kweSmoothQuant kuchiwedzera zvakanyanya pasina nguva yekumhanya?
   • Inomhanya pane imwe yakatsaurirwa GPU
   • Inongoshanda panguva yekudzidziswa, isingambofungidzire
   • Mazuva ano maCPU anofuratira kuyera zvinhu
   • Zvikero zvinopinzwa kunze kweindaneti muhuremu hwepedyo kana mashandiro akasanganiswa
5. Ndeipi basa rinoitwa nealpha hyperparameter muSmoothQuant?
   • Iyo inodzora kuti wakawanda sei mutoro wekunze unochinja kubva ku activation kuenda kune huremu
   • Inoisa mwero wekudzidza wekugadzirisa zvakanaka
   • Inotarisa nhamba ye quantization bits
   • Inosarudza kuti ndezvipi zvinotungamira kuchekerera
6. What result did the SmoothQuant paper report for its evaluated W8A8 large-language-model deployments?
   • A guaranteed 10× speedup on every GPU
   • About 1.56× speedup and roughly 2× memory savings with near-full-precision accuracy
   • No runtime or memory benefit
   • A 50% accuracy drop in every tested model
7. Why does multiplying activation channels by 1/s and corresponding weights by s preserve a linear layer’s product?
   • The output is rounded to the nearest class
   • The model replaces multiplication with addition
   • The reciprocal scale factors cancel in the activation-weight multiplication
   • The weights are set to zero after scaling
8. In SmoothQuant’s scale formula, what does increasing alpha generally do to outlier handling?
   • It removes the model’s weight matrices
   • It turns W8A8 into W16A16
   • It eliminates all quantization error
   • It shifts more scaling burden away from activations and onto weights