SkyRL Megatron Development Tracker H1 2026 #1392

This issue tracks the development of the megatron backend for SkyRL for large scale MoE training.

Megatron-Core MoE tech report: https://arxiv.org/pdf/2603.07685
Prior tracking issue: #203

New Features

• [P0] Support R3 - Rollout Routing Replay ([skyrl-train] Enable routing replay in SkyRL #815)
• [P2] Megatron dynamic context parallel support ([megatron][perf] Integrate megatron dynamic context parallelism #1019)
• [P2] Enabling TransformerConfig.cp_comm_type="a2a" for ulysses style sequence parallelism in megatron [megatron] compute logprobs/entropy on CP sharded logits + fix CP loss scaling #1751
• [P2] Support Virtual Pipeline Parallel for improving training throughput - marked as not planned for now
• [P2] Support megatron native FSDP
• [P1] Enable dynamic batch sizing in megatron feat: add max_tokens_per_microbatch config for token-based micro-batching #1477

Megatron + LoRA

• [P1] Support in memory LoRA only weight sync for megatron + lora ([train] Support LoRA-only weight syncing for Megatron backend #1336)
• [P1] LoRA rank normalization via Megatron-Bridge ([feature] LoRA rank normalization for MoE models NVIDIA-NeMo/Megatron-Bridge#2964)
• [P0] Improve LoRA checkpointing support and fix/note upstream bugs (Dense LoRA adapters lose TP shards during PEFT-filtered checkpoint save NVIDIA-NeMo/Megatron-Bridge#2240)

Bugs/Improvements

• [P0] Add a DAPO recipe with MoE + megatron to CI: [CI] Add a DAPO recipe with the Megatron backend to CI #1322
• [P1] Debug GLM + LoRA train/infer mismatch issue