
Guides selective and full activation recompute in Megatron Bridge to cut GPU memory use
- Open source
- 4K GitHub stars
- AI models and LLM tooling


Guides selective and full activation recompute in Megatron Bridge to cut GPU memory use

Guides an agent through reducing peak GPU memory in Megatron Bridge training runs

Picks Megatron Bridge training recipes and resizes parallelism for your GPU allocation

Guides choosing and combining TP, DP, PP, CP and EP degrees for Megatron Bridge training

Guides long-context MoE training in Megatron Bridge: CP sizing, recompute and dispatcher

Guides tuning of MoE expert-parallel communication overlap in Megatron Bridge

Hardware-specific MoE training playbooks and candidate parallelism configs for Megatron

Runs Megatron-LM and Megatron Bridge training and checks their loss curves match

Guides an agent through enabling Megatron FSDP sharded training in Megatron-Bridge

Guides an agent through enabling hierarchical context parallelism in Megatron-Bridge

Guides enabling and verifying TP, DP and PP communication overlap in Megatron-Bridge

Guides Megatron-Bridge sequence packing and long-context training setup and validation
The closest alternatives to Nemo mbridge perf cpu offloading are Nemo mbridge perf activation recompute, Nemo mbridge perf memory tuning, Nemo mbridge recipe recommender, Nemo mbridge perf parallelism strategies and Nemo mbridge perf moe long context.
Yes. Nemo mbridge perf activation recompute, Nemo mbridge perf memory tuning, Nemo mbridge recipe recommender have a free or open-source option.