Skip to main content
skill CX Codex

nemo-mbridge-perf-cuda-graphs — Claude Code skill for CUDA graphs in Megatron Bridge

nemo-mbridge-perf-cuda-graphs is a skill that configures CUDA graph capture in Megatron Bridge with local full-iteration or Transformer Engine scoped graphs. It can reduce host-driver overhead or help trace a crash to a graph config change.

Apache-2.0 1,900 installs ★ 4,106 updated 2026-10-23

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add NVIDIA/skills --skill nemo-mbridge-perf-cuda-graphs -a claude-code
Open on GitHub

What nemo-mbridge-perf-cuda-graphs does

CUDA graphs capture GPU work once and replay it with little host-driver overhead. Bridge has two paths: "local" wraps the full forward and backward with MCore's FullCudaGraphWrapper, while "transformer_engine" applies TE make_graphed_callables() per layer for scopes like attn, mlp, moe_router, or mamba.

The skill gives a clear starting point. Most workloads should begin with TE scoped graphs: attn for dense models, attn moe_router moe_preprocess for dropless MoE. Save local with full_iteration for cases where you genuinely need whole-iteration capture and accept the tighter constraints.

Constraints matter here. use_te_rng_tracker is enforced, full_iteration demands check_for_nan_in_loss = False, moe and moe_router scopes cannot be combined, tensor shapes must stay static, and CPU offloading breaks graphs. With PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True you also need NCCL_GRAPH_REGISTER=0. The bring-up order is sensible: stabilise eager first, fix sequence length and micro-batch size, enable the narrowest scope, then compare replay against eager after warmup and capture.

What it helps with

  • reduce host-driver overhead with CUDA graphs
  • choose between local and Transformer Engine graph implementations
  • write a working config for a dense model
  • set the right scope for dropless MoE training
  • debug a regression after changing CUDA graph settings

Example prompts for the agent

  • Set up CUDA graphs for a dropless MoE model, starting with a narrow scope
  • List the constraints for local full_iteration graphs in Megatron Bridge
  • Walk me through enabling CUDA graphs after a stable eager run

How to install nemo-mbridge-perf-cuda-graphs

1Skills CLI (Vercel)

npx skills add NVIDIA/skills --skill nemo-mbridge-perf-cuda-graphs -a claude-code

2Manually: put the files in place

~/.claude/skills/nemo-mbridge-perf-cuda-graphs/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/nemo-mbridge-perf-cuda-graphs/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about nemo-mbridge-perf-cuda-graphs

How do I install nemo-mbridge-perf-cuda-graphs in Claude Code?

Run npx skills add NVIDIA/skills --skill nemo-mbridge-perf-cuda-graphs -a claude-code. The skill lands at ~/.claude/skills/nemo-mbridge-perf-cuda-graphs/SKILL.md.

What does this skill do?

It helps enable and validate CUDA graphs in Megatron Bridge. It covers both implementations, their constraints, and a practical bring-up order.

Why can't TE scoped graphs capture MoE modules with dropless routing?

Token-dropless MoE routing limits graphable scope to dense modules. That's why the recommended scope is attn moe_router moe_preprocess, not moe itself.

Source text

SKILL.md · Apache-2.0
---
name: nemo-mbridge-perf-cuda-graphs
description: Validate and use CUDA graph capture in Megatron Bridge, including local full-iteration graphs and Transformer Engine scoped graphs for attention, MLP, and MoE modules.
license: Apache-2.0
when_to_use: Reducing host-driver overhead via CUDA graphs, or tracing a crash or regression to a CUDA graph config change; 'cuda_graph_impl', 'full iteration graph', 'TE scoped graph', 'graphed callables', 'CUDA graph capture'.
---

# CUDA Graphs

Stable documentation: @docs/training/cuda-graphs.md
Card: @skills/nemo-mbridge-perf-cuda-graphs/card.yaml

<!-- NVSkills CI refresh: 2026-06-15. No instruction changes. -->

## What It Is

CUDA graphs capture GPU operations once and replay them with minimal
host-driver overhead. Bridge supports two implementations:

| `cuda_graph_impl` | Mechanism | Scope support |
|---|---|---|
| `"local"` | MCore `FullCudaGraphWrapper` wrapping entire fwd+bwd | `full_iteration` |
| `"transformer_engine"` | TE `make_graphed_callables()` per layer | `attn`, `mlp`, `moe`, `moe_router`, `moe_preprocess`, `mamba` |

## Quick Decision

Start with TE-scoped graphs for most training workloads, then verify replay
timing against eager on the same dispatcher, layout, and container:

- dense models: `attn`, then optionally `mlp`
- dropless MoE: `attn moe_router moe_preprocess`
- VLMs: the same dropless-MoE scope, but only after the real-data path is stable

Use `local` + `full_iteration` only when you specifically want full-iteration
capture and can satisfy the tighter constraints.

For recompute-heavy workloads:

- TE-scoped graphs pair naturally with selective recompute
- full recompute usually pushes you toward `local` full-iteration graphs or away
  from graphs entirely

Related docs:

- @docs/training/cuda-graphs.md
- @docs/training/activation-recomputation.md

## Enablement

### Local full-iteration graph

```python
cfg.model.cuda_graph_impl = "local"
cfg.model.cuda_graph_scope = ["full_iteration"]
cfg.model.cuda_graph_warmup_steps = 3
cfg.model.use_te_rng_tracker = True
cfg.rng.te_rng_tracker = True
cfg.rerun_state_machine.check_for_nan_in_loss = False
cfg.ddp.check_for_nan_in_grad = False
```

### TE scoped graph (dense model)

```python
cfg.model.cuda_graph_impl = "transformer_engine"
cfg.model.cuda_graph_scope = ["attn"]           # or ["attn", "mlp"]
cfg.model.cuda_graph_warmup_steps = 3
cfg.model.use_te_rng_tracker = True
cfg.rng.te_rng_tracker = True
```

### TE scoped graph (MoE model)

```python
cfg.model.cuda_graph_impl = "transformer_engine"
cfg.model.cuda_graph_scope = ["attn", "moe_router", "moe_preprocess"]
cfg.model.cuda_graph_warmup_steps = 3
cfg.model.use_te_rng_tracker = True
cfg.rng.te_rng_tracker = True
```

### Performance harness CLI

```bash
uv run python scripts/performance/run_script.py \
  -m qwen \
  -mr qwen3_30b_a3b \
  --task pretrain \
  -g h100 \
  -c bf16 \
  -ng 16 \
  --cuda_graph_impl transformer_engine \
  --cuda_graph_scope attn,moe_router,moe_preprocess \
  ...
```

Valid CLI values live in `scripts/performance/argument_parser.py`:
- `VALID_CUDA_GRAPH_IMPLS`: `["none", "local", "transformer_engine"]`
- `VALID_CUDA_GRAPH_SCOPES`: `["full_iteration", "attn", "mlp", "moe", "moe_router", "moe_preprocess", "mamba"]`

The performance harness uses a comma-separated `--cuda_graph_scope` value and
auto-enables `model.use_te_rng_tracker` plus `rng.te_rng_tracker` when
`--cuda_graph_impl` is not `none`.

### Required constraints

- `use_te_rng_tracker = True` (enforced in `gpt_provider.py`)
- `full_iteration` scope only with `cuda_graph_impl = "local"`
- `full_iteration` scope requires `check_for_nan_in_loss = False`
- Do not combine `moe` scope and `moe_router` scope
- Tensor shapes must be static (fixed seq_length, fixed micro_batch_size)
- MoE token-dropless routing limits graphable scope to dense modules
- With `PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True`, set
  `NCCL_GRAPH_REGISTER=0` (MCore enforces for local impl on arch < sm_100;
  TE impl asserts unconditionally)
- CPU offloading is incompatible with CUDA graphs
- `moe_preprocess` scope requires `moe_router` scope to also be set

### Practical bring-up order

1. Stabilize the eager run first.
2. Fix sequence length and micro-batch size.
3. Enable the narrowest useful graph scope.
4. Confirm replay is active and memory is still acceptable.
5. Compare eager against graph replay iterations after warmup and capture; do
   not include the capture step in steady-state timing.
6. Only then widen scope or combine with overlap features.

## Code Anchors

### Bridge config and validation

```1524:1531:src/megatron/bridge/training/config.py
        # CUDA graph scope validation: check_for_nan_in_loss must be disabled with full_iteration graph
        if self.model.cuda_graph_impl == "local" and CudaGraphScope.full_iteration in self.model.cuda_graph_scope:
            assert not self.rerun_state_machine.check_for_nan_in_loss, (
                "check_for_nan_in_loss must be disabled when using full_iteration CUDA graph. "
                "Set rerun_state_machine.check_for_nan_in_loss=False."
            )
        if self.model.cuda_graph_impl == "none":
            self.model.cuda_graph_scope = []
```

### TE RNG tracker requirement

```213:216:src/megatron/bridge/models/gpt_provider.py
        if self.cuda_graph_impl != "none":
            assert getattr(self, "use_te_rng_tracker", False), (
                "Transformer engine's RNG tracker is required for cudagraphs, it can be "
                "enabled with use_te_rng_tracker=True'."
```

### Graph creation and capture in training loop

```231:255:src/megatron/bridge/training/train.py
    # Capture CUDA Graphs.
    cuda_graph_helper = None
    if model_config.cuda_graph_impl == "transformer_engine":
        cuda_graph_helper = TECudaGraphHelper(...)
    # ...
    if config.model.cuda_graph_impl == "local" and CudaGraphScope.full_iteration in config.model.cuda_graph_scope:
        forward_backward_func = FullCudaGraphWrapper(
            forward_backward_func, cuda_graph_warmup_steps=config.model.cuda_graph_warmup_steps
        )
```

### TE graph capture after warmup

```338:350:src/megatron/bridge/training/train.py
        # Capture CUDA Graphs after warmup.
        if (
            model_config.cuda_graph_impl == "transformer_engine"
            and cuda_graph_helper is not None
            and not cuda_graph_helper.graphs_created()
            and global_state.train_state.step - start_iteration == model_config.cuda_graph_warmup_steps
        ):
            if model_config.cuda_graph_warmup_steps > 0 and should_toggle_forward_pre_hook:
                disable_forward_pre_hook(model, param_sync=False)
            cuda_graph_helper.create_cudagraphs()
            if model_config.cuda_graph_warmup_steps > 0 and should_toggle_forward_pre_hook:
                enable_forward_pre_hook(model)
                cuda_graph_helper.cuda_graph_set_manual_hooks()
```

### RNG initialization

```199:206:src/megatron/bridge/training/initialize.py
        _set_random_seed(
            rng_config.seed,
            rng_config.data_parallel_random_init,
            rng_config.te_rng_tracker,
            rng_config.inference_rng_tracker,
            use_cudagraphable_rng=(model_config.cuda_graph_impl != "none"),
            pg_collection=pg_collection,
        )
```

### Delayed wgrad + CUDA graph interaction

```522:555:src/megatron/bridge/training/comm_overlap.py
            cuda_graph_scope = getattr(model_cfg, "cuda_graph_scope", []) or []
            # ... scope parsing ...
            if wgrad_in_graph_scope:
                assert is_te_min_version("2.12.0"), ...
                assert model_cfg.gradient_accumulation_fusion, ...
                if attn_scope_enabled:
                    assert not model_cfg.add_bias_linear and not model_cfg.add_qkv_bias, ...
```

### Perf harness override helper

```102:124:scripts/performance/utils/overrides.py
def _set_cuda_graph_overrides(
    recipe, cuda_graph_impl=None, cuda_graph_scope=None
):
    # Sets impl, scope, and auto-enables te_rng_tracker
```

### Graph cleanup

```1414:1441:src/megatron/bridge/training/train.py
def _delete_cuda_graphs(cuda_graph_helper):
    # Deletes FullCudaGraphWrapper and TE graph objects to free NCCL buffers
```

### MCore classes (in 3rdparty/Megatron-LM)

- `CudaGraphManager`: `megatron/core/transformer/cuda_graphs.py`
- `TECudaGraphHelper`: `megatron/core/transformer/cuda_graphs.py`
- `FullCudaGraphWrapper`: `megatron/core/full_cuda_graph.py`
- `CudaGraphScope` enum: `megatron/core/transformer/enums.py`

### Positive recipe anchors

- `src/megatron/bridge/perf_recipes/deepseek/gb300/deepseek_v3.py`
- `src/megatron/bridge/perf_recipes/qwen/gb300/qwen3_moe.py`
- `src/megatron/bridge/perf_recipes/gpt_oss/gb300/gpt_oss.py`

### Tests

| File | Coverage |
|---|---|
| `tests/unit_tests/training/test_config.py` | `full_iteration` NaN-check constraint |
| `tests/unit_tests/training/test_comm_overlap.py` | `delay_wgrad` + CUDA graph interaction |
| `tests/unit_tests/models/test_gpt_full_te_layer_autocast_spec.py` | TE autocast with CUDA graphs |
| `tests/functional_tests/test_groups/recipes/test_llama_recipes_pretrain_cuda_graphs.py` | End-to-end local and TE graph smoke tests |
| `tests/unit_tests/recipes/kimi/test_kimi_k2.py` | TE + CUDA graph recipe config |
| `tests/unit_tests/recipes/gpt/test_gpt3_175b.py` | TE + CUDA graph recipe config |
| `tests/unit_tests/recipes/qwen_vl/test_qwen25_vl_recipes.py` | VLM CUDA graph settings |

## Pitfalls

1. **TE RNG tracker is mandatory**: Setting `cuda_graph_impl` without
   `use_te_rng_tracker=True` and `rng.te_rng_tracker=True` will assert
   in the provider.

2. **`full_iteration` requires NaN checks disabled**: The entire fwd+bwd is
   captured, so loss-NaN checking cannot inspect intermediate values.

3. **MoE scope restrictions**: `moe` scope and `moe_router` scope are
   mutually exclusive. Token-dropless MoE can only graph `moe_router` and
   `moe_preprocess`, not the full expert dispatch.

4. **Memory overhead**: CUDA graphs pin all intermediate buffers for the
   graph's lifetime (no memory reuse). TE scoped graphs add a few GB;
   full-iteration graphs can increase peak memory by 1.5–2×. `PP > 1`
   compounds overhead since each stage holds its own graph.

5. **Delayed wgrad interaction**: When `delay_wgrad_compute=True` and
   attention or MoE router is in `cuda_graph_scope`, additional constraints
   apply: TE >= 2.12.0, `gradient_accumulation_fusion=True`, and no
   attention bias.

6. **Variable-length sequences break graphs**: Sequence lengths must be
   constant across steps. Use padded packed sequences if packing is needed.

7. **Graph cleanup is required**: CUDA graph objects hold NCCL buffer
   references. Bridge handles this in `_delete_cuda_graphs()` at the end
   of training, but early exits must call it explicitly.

8. **Older GPU architectures**: On GPUs with compute capability < 10.0
   (pre-Blackwell), set `NCCL_GRAPH_REGISTER=0` when using
   `PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True`. Enforced in MCore
   `CudaGraphManager` (cuda_graphs.py:1428) and `TECudaGraphHelper`
   (cuda_graphs.py:1697). The TE impl asserts unconditionally regardless
   of arch.

9. **CPU offloading incompatible**: CUDA graphs cannot be used with CPU
   offloading. Enforced in MCore `transformer_config.py:1907`.

10. **MoE recompute + moe_router scope**: MoE recompute is not supported
    with `moe_router` CUDA graph scope when using `cuda_graph_impl =
    "transformer_engine"`. Enforced in MCore `transformer_config.py:1977`.

11. **Layer-level recompute requires `full_iteration` scope**: Using
    `recompute_granularity="full"` with `recompute_num_layers` (recompute N
    whole transformer layers) is incompatible with TE-scoped graphs. MCore
    calls this "full" granularity even though you're selecting how many
    layers — the name refers to recomputing the full layer, not full model.
    Any TE-scoped scope (`attn`, `mlp`, `moe_router`, etc.) will assert:
    `AssertionError: full recompute is only supported with full iteration CUDA graph.`
    This commonly hits FP8 configs that default to TE-scoped graphs (e.g.
    `LLAMA3_70B_SFT_CONFIG_H100_FP8_CS_V1` uses `cuda_graph_impl=
    "transformer_engine"`, `cuda_graph_scope="mlp"`). Fix: use submodule
    recompute (`recompute_granularity="selective"` + `recompute_modules`),
    disable CUDA graphs, or switch to `local` + `full_iteration`. Enforced
    in MCore `transformer_config.py:2001-2005`. See also
    @skills/nemo-mbridge-perf-activation-recompute/SKILL.md.

12. **Benchmark numbers are workload-specific**: graph wins are usually real
    when host overhead is visible, but the exact gain depends on batch shape,
    PP depth, recompute, dispatcher backend, and whether the eager baseline was
    already optimized.

13. **A successful capture is not a speedup guarantee**: On 2026-05-18,
    Qwen3 30B A3B H100 BF16 pretrain with the all-to-all dispatcher captured
    TE-scoped `attn,moe_router,moe_preprocess` graphs successfully (`48`
    graphable layers, about `6.9 s` capture time on rank 0), but replay
    iterations 5-8 averaged `42.00 s` versus `41.36 s` for eager. Treat
    scoped graphs as a bring-up candidate and validate on the target stack.

## Verification

### Unit tests

```bash
uv run python -m pytest \
  tests/unit_tests/training/test_config.py -k "cuda_graph" \
  tests/unit_tests/training/test_comm_overlap.py -k "cuda_graph" \
  tests/unit_tests/models/test_gpt_full_te_layer_autocast_spec.py -k "cuda_graph" -q
```

### Functional smoke test (requires GPU)

```bash
uv run python -m pytest \
  tests/functional_tests/test_groups/recipes/test_llama_recipes_pretrain_cuda_graphs.py -q
```

### Success criteria

- Unit tests pass, covering config validation for both `local` and
  `transformer_engine` implementations.
- Functional test completes training steps with both CUDA graph
  implementations.
- No NCCL errors or illegal memory access in logs.

© NVIDIA · Apache-2.0 · NVIDIA/skills

Similar skills

All skills for AI and ML →

From the same set312

The whole set →
dynamo-troubleshoot dynamo-troubleshoot is a Claude Code skill that investigates unhealthy Dynamo deployments. It activates for failures in pods, PVCs, model-cache jobs, workers, endpoints, or benchmarks. CX Codex +1 1,900installs nemo-mbridge-recipe-recommender nemo-mbridge-recipe-recommender — a skill for Claude Code selects and tunes Megatron Bridge training recipes based on model architecture, GPU count, and workload. Configures TP, CP, and SP layout for pretraining, SFT, or PEFT. CX Codex +1 1,900installs launch-nemo-rl launch-nemo-rl is a Claude Code skill for launching, monitoring, and debugging NeMo-RL recipes on Kubernetes with the nrl-k8s CLI. CX Codex +1 1,900installs nemo-mbridge-perf-memory-tuning nemo-mbridge-perf-memory-tuning is a Claude Code skill for lowering GPU memory peaks and diagnosing OOM during Megatron Bridge training. It covers allocator settings, parallelism, and activation memory. CX Codex +1 1,900installs nemo-rl-auto-research nemo-rl-auto-research automates hypothesis testing and training loops in NeMo-RL and NeMo-gym. Keeps experiments tracked through dedicated git branches and TSV logs. CX Codex +1 1,900installs nemo-automodel-launcher-config nemo-automodel-launcher-config is a Claude Code skill for configuring NeMo AutoModel launches with torchrun, Slurm, or SkyPilot. It activates for cluster submissions, multi-node scripts, and launch failures. CX Codex +1 1,900installs earth2studio-deterministic-forecast earth2studio-deterministic-forecast — a skill that walks you through building a deterministic weather forecast script with Earth2Studio: model, data source, IO backend and inference. CX Codex +1 1,900installs earth2studio-data-fetch earth2studio-data-fetch maps a weather data request onto Earth2Studio sources: checks the lexicon for variable support and writes a Python script returning an xarray DataArray. CX Codex +1 1,900installs

Project sponsors

Companies that keep this analytics open