@inproceedings{14461,
  abstract     = {Communication-reduction techniques are a popular way to improve scalability in data-parallel training of deep neural networks (DNNs). The recent emergence of large language models such as GPT has created the need for new approaches to exploit data-parallelism. Among these, fully-sharded data parallel (FSDP) training is highly popular, yet it still encounters scalability bottlenecks. One reason is that applying compression techniques to FSDP is challenging: as the vast majority of the communication involves the model’s weights, direct compression alters convergence and leads to accuracy loss. We present QSDP, a variant of FSDP which supports both gradient and weight quantization with theoretical guarantees, is simple to implement and has essentially no overheads. To derive QSDP we prove that a natural modification of SGD achieves convergence even when we only maintain quantized weights, and thus the domain over which we train consists of quantized points and is, therefore, highly non-convex. We validate this approach by training GPT-family models with up to 1.3 billion parameters on a multi-node cluster. Experiments show that QSDP preserves model accuracy, while completely removing the communication bottlenecks of FSDP, providing end-to-end speedups of up to 2.2x.},
  author       = {Markov, Ilia and Vladu, Adrian and Guo, Qi and Alistarh, Dan-Adrian},
  booktitle    = {Proceedings of the 40th International Conference on Machine Learning},
  issn         = {2640-3498},
  location     = {Honolulu, Hawaii, HI, United States},
  pages        = {24020--24044},
  publisher    = {ML Research Press},
  title        = {{Quantized distributed training of large models with convergence guarantees}},
  volume       = {202},
  year         = {2023},
}