Distributed Training

Data, tensor and pipeline parallelism, ZeRO/FSDP sharding, mixed precision, and offload.

13concepts
165flashcards
111minutes of reading
  1. 01 Data Parallelism and DDP How replicating the model and sharding the batch across GPUs scales training, and why AllReduce is the primitive every framework eventually depends on. intermediate 8m 5 cards
  2. 02 Fault Tolerance at Scale Why a 16,000-GPU run fails every few hours, how checkpoint interval trades wasted compute against write cost, and what asynchronous and in-memory checkpointing changed. intermediate 8m 15 cards
  3. 03 Gradient Checkpointing, Activation Recomputation, and CPU Offload Why activations - not weights - usually dominate training memory, and how recomputation and CPU/NVMe offload trade compute and bandwidth to fit larger models. intermediate 9m 5 cards
  4. 04 Mixed-Precision Training (FP16, BF16, FP8) How lower-precision formats halve memory and double throughput on tensor cores, why BF16 displaced FP16 for training, and what FP8 changes on H100 and Blackwell. intermediate 8m 4 cards
  5. 05 Composing Parallelism Strategies How data, tensor, pipeline, context and expert parallelism combine into one device mesh, why the ordering follows the interconnect hierarchy, and what each axis costs. advanced 9m 12 cards
  6. 06 Expert Parallelism and All-to-All Sparse mixture-of-experts models are sharded by placing different experts on different devices, which replaces the all-reduce of dense training with two all-to-all exchanges per layer and makes routing a network-topology problem. advanced 9m 24 cards
  7. 07 Gradient Compression and Quantised Collectives Ninety-nine point nine percent of the values in a distributed SGD gradient exchange are redundant, and twenty years of compression research shows that exploiting this is easy in theory and constrained in practice by one property: whether the compressed form survives an all-reduce. advanced 8m 24 cards
  8. 08 Low-Communication Distributed Training Standard data parallelism synchronises gradients every step, which requires a datacentre-grade fabric; DiLoCo synchronises every few hundred steps instead and trains language models across poorly connected islands of accelerators. advanced 9m 24 cards
  9. 09 Overlapping Communication and Computation Why a large training run's collectives are mostly free when overlapped and catastrophic when exposed, and the bucketing, prefetch and scheduling tricks that hide them. advanced 8m 15 cards
  10. 10 Pipeline Bubbles and Schedules Why pipeline parallelism wastes device time by construction, how the bubble fraction depends on micro-batch count, and what 1F1B, interleaving and zero-bubble schedules recover. advanced 8m 12 cards
  11. 11 Sequence and Context Parallelism Why long-context training runs out of memory even when the model fits, and how Ring Attention and Ulysses split the sequence dimension across devices without changing the maths. advanced 8m 15 cards
  12. 12 Tensor and Pipeline Parallelism How frontier labs split a model across thousands of GPUs by sharding within layers (tensor parallel) and across layers (pipeline parallel), and how to pick the split. advanced 10m 5 cards
  13. 13 ZeRO and FSDP How sharding optimiser state, gradients, and parameters across data-parallel ranks turns a memory problem into a bandwidth problem, and why FSDP is now the PyTorch default. advanced 9m 5 cards