
Guides an agent through reducing peak GPU memory in Megatron Bridge training runs
- Open source
- 4K GitHub stars
- Knowledge and memory


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

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

Configures CPU offloading in Megatron Bridge to cut GPU memory during training

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

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

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

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

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

Guides choice of MoE token dispatcher (alltoall, DeepEP, HybridEP) by hardware and EP size

Guidance for choosing FSDP or 3D parallelism when training MoE vision-language models

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

Guides an agent through enabling Megatron FSDP sharded training in Megatron-Bridge
The closest alternatives to Nemo mbridge perf activation recompute are Nemo mbridge perf memory tuning, Nemo mbridge perf moe long context, Nemo mbridge perf cpu offloading, Nemo mbridge perf moe hardware configs and Nemo mbridge recipe recommender.
Yes. Nemo mbridge perf memory tuning, Nemo mbridge perf moe long context, Nemo mbridge perf cpu offloading have a free or open-source option.