Kuiz berkaitan panduan · Sederhana Tahap

Kuiz Latihan Ketepatan Campuran

Meliputi julat FP16 dan penskalaan kerugian, bfloat16, pemberat induk, autocast, pengumpulan Teras Tensor dan keuntungan bergantung kepada perkakasan yang diukur.

Laluan panduan yang berkaitanLatihan Ketepatan Campuran
soalan 1 daripada 8

In an FP16 training recipe that keeps an FP32 master weight, why retain it?

Pratonton setiap soalan8 soalan

Jawapan dan penjelasan muncul semasa anda mengambil kuiz.

  1. In an FP16 training recipe that keeps an FP32 master weight, why retain it?
    • Untuk mengekalkan ketepatan dalam kemas kini berat kecil yang boleh dibundarkan 16-bit
    • Untuk menjadikan fail model lebih kecil pada cakera
    • Untuk membenarkan latihan tanpa GPU
    • Untuk menyulitkan parameter model
  2. Apakah masalah yang diselesaikan oleh penskalaan kerugian dalam latihan FP16?
    • Model overfitting data
    • Kecerunan mengalir ke sifar kerana julat FP16 adalah sempit
    • GPU kehabisan ruang cakera
    • Pemuatan data perlahan dari cakera
  3. Apakah kelebihan bfloat16 berbanding FP16?
    • Ia menggunakan lebih sedikit jumlah bit
    • Ia menyimpan teks dan bukannya nombor
    • Ia mempunyai julat eksponen luas yang sama seperti FP32, mengurangkan keperluan untuk penskalaan kerugian
    • Ia hanya berjalan pada CPU
  4. For many supported Tensor Core matrix operations, how are partial sums commonly accumulated?
    • In integer arithmetic with no rounding
    • In text before converting to FP16
    • In FP16 only, which prevents every numerical error
    • In FP32, which reduces rounding error compared with low-precision accumulation
  5. What is a common benefit of using 16-bit values for supported training operations?
    • They can reduce tensor memory and improve throughput, with gains depending on hardware and workload
    • They guarantee higher model accuracy
    • They remove the need to compute gradients
    • They always double end-to-end training speed
  6. Which current PyTorch APIs are commonly used for automatic mixed-precision training?
    • torch.cuda.amp only, which is the recommended new namespace
    • torch.amp.autocast and, for FP16 gradient scaling, torch.amp.GradScaler
    • torch.load and torch.save
    • torch.nn.DataParallel and torch.manual_seed
  7. Why does bfloat16 often need less loss scaling than FP16?
    • It has more mantissa bits than FP32
    • It stores gradients as integers
    • It has an exponent range like FP32, so very small gradients are less likely to underflow
    • It never rounds values
  8. What does autocast do in automatic mixed precision?
    • Converts every tensor and operation to FP16 unconditionally
    • Replaces the optimizer with a different algorithm
    • Guarantees identical bitwise results to FP32
    • Selects operation dtypes so supported operations can use lower precision while others retain safer precision