Kuis terkait panduan · Sedang Tingkat
Kuis Pemeriksaan Gradien
Menjelaskan rematerialisasi aktivasi, penghematan memori, overhead komputasi yang bergantung pada jadwal, dan kombinasi dengan sharding dan presisi campuran.
Pertanyaan 1 dari 8
Apa yang terutama diperdagangkan oleh pos pemeriksaan gradien untuk menghemat memori?
Pratinjau setiap pertanyaan8 pertanyaan
Jawaban dan penjelasan muncul saat Anda mengikuti kuis.
- Apa yang terutama diperdagangkan oleh pos pemeriksaan gradien untuk menghemat memori?
- Komputasi ekstra (komputasi ulang) untuk penggunaan memori yang lebih rendah
- Akurasi model untuk kecepatan
- Ruang disk untuk bandwidth jaringan
- Presisi untuk ukuran batch
- Mengapa aktivasi biasanya disimpan selama forward pass?
- Untuk menampilkannya kepada pengguna
- Karena propagasi mundur memerlukannya untuk menghitung gradien
- Untuk mempercepat pemuatan data
- Mereka sebenarnya tidak dibutuhkan
- In a simplified chain of N similarly sized layers, what can checkpoint boundaries spaced about sqrt(N) layers apart achieve in the textbook schedule?
- It stores no activations and performs no recomputation
- It guarantees constant memory for every model graph
- Activation storage can scale on the order of sqrt(N) checkpoints
- It makes optimizer-state memory scale as sqrt(N)
- What most determines gradient-checkpointing runtime overhead in a practical model?
- Only the model’s file extension
- A fixed 20–30% slowdown for every model
- The number of labels in the dataset
- Which operations are recomputed, checkpoint placement, graph structure, and hardware
- Di PyTorch, utilitas mana yang biasa digunakan untuk menerapkan titik pemeriksaan gradien ke modul?
- torch.utils.checkpoint
- obor.optim.SGD
- obor.nn.Putus sekolah
- torch.cuda.empty_cache
- What happens to operations inside a checkpointed region during backward?
- Their gradients are skipped entirely
- Their forward computations are rerun to regenerate needed activations
- Their weights are permanently discarded
- They are executed only during inference
- Which training techniques can be combined with activation checkpointing to reduce different memory costs?
- Only increasing the number of labels
- Turning off backpropagation
- FSDP for sharded model state and mixed precision for lower-precision tensors
- Replacing the model with a tokenizer
- What is selective activation checkpointing intended to choose?
- Which training labels are removed
- Which model outputs are published
- Which GPU becomes the tokenizer
- Which operations to save and which to recompute based on their memory and runtime costs