训练与算法
==================================================

理解训练目标、精度、并行方式与显存。

.. toctree::
   :maxdepth: 1

   ../advanced/policy-mismatch.md
   ../advanced/parallelism-memory.md
   ../advanced/low-precision.md
   ../advanced/reproducibility.md
   ../advanced/on-policy-distillation.md
   ../advanced/megatron-config.md
   ../advanced/arch-support-beyond-megatron.md