🧰 Megatron-Core

The items that name it 3

• Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs
  article2026-10-01

  NVIDIA's established framework/option for training large MoE models, compared against as an alternative to Olmo-core 3's integrated MoE training stack.

• baidu-baige/LoongSage
  repo2026-08-28

  Training backend for LoongSage, providing core capabilities for large-scale distributed training

• Granite 4.2 LLMs: How They're Built
  article2026-08-25

  Training backend used by NeMo-RL for the GRPO training loop.