Nemo mbridge perf megatron fsdp vs Nemo mbridge recipe recommender
Nemo mbridge perf megatron fsdp is an open-source agent skill for ai models and llm tooling. Nemo mbridge recipe recommender is an open-source agent skill for ai models and llm tooling.
- Nemo mbridge perf megatron fsdp is best for switch a megatron-bridge pretraining recipe from ddp to fsdp data parallelism.
- Nemo mbridge recipe recommender is best for pick a starting recipe for pretraining an 8b model on one node.
- What it is
- Nemo mbridge perf megatron fsdp is an open-source agent skill for ai models and llm tooling.Nemo mbridge recipe recommender is an open-source agent skill for ai models and llm tooling.
- Best for
- Switch a Megatron-Bridge pretraining recipe from DDP to FSDP data parallelismPick a starting recipe for pretraining an 8B model on one node
- Kind
- SkillSkill
- Pricing
- Open sourceOpen source
- Platforms
- Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf, GooseClaude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf, Goose
- GitHub stars
- 4K4K
- Last commit
- Nov 2026Nov 2026
- License
- Apache-2.0Apache-2.0
- Open source
- Good for
- Switch a Megatron-Bridge pretraining recipe from DDP to FSDP data parallelism; Fix checkpoint save and load errors after enabling Megatron FSDP; Trace an out-of-memory error or regression to an FSDP config changePick a starting recipe for pretraining an 8B model on one node; Resize a 70B recipe to fit a 32-GPU allocation; Set up a LoRA fine-tune on a single GPU
What Nemo mbridge perf megatron fsdp does
FSDP config overrides
Shows the DDP and checkpoint settings needed to switch a recipe to Megatron FSDP.
Recipe fixup example
Demonstrates applying the overrides to an existing Llama 3 8B pretraining recipe.
Code anchor references
Points to the config validation and runtime wrapper selection in the Bridge source.
Known pitfalls list
Covers checkpoint format defaults, FSDP2 conflicts, offloading limits and CUDA settings.
Smoke test verification
Gives a 2-GPU pytest command and the success criteria to confirm the setup works.
What Nemo mbridge recipe recommender does
Recipe lookup by goal
Maps a model size, training mode and GPU budget to a suitable library or benchmark recipe name.
Parallelism resizing rules
Explains constraints for tensor, pipeline, context and expert parallelism when changing GPU counts.
Convergence vs. execution changes
Separates overrides that change training outcomes from those that only change hardware mapping.
Benchmark versus library split
Distinguishes throughput-only benchmark recipes from functional training recipes and their launchers.
Common pitfalls table
Lists frequent errors such as TP divisibility, CUDA graph conflicts and MoE memory issues with fixes.
Nemo mbridge perf megatron fsdp vs Nemo mbridge recipe recommender: questions
Is Nemo mbridge perf megatron fsdp or Nemo mbridge recipe recommender cheaper?
Both are open source. Compare current plans on each official site.
What is the difference between Nemo mbridge perf megatron fsdp and Nemo mbridge recipe recommender?
Nemo mbridge perf megatron fsdp is an open-source agent skill for ai models and llm tooling. Nemo mbridge recipe recommender is an open-source agent skill for ai models and llm tooling.
