Distributed Training in NeMo AutoModel

Purpose

NeMo AutoModel uses PyTorch-native distributed training. All parallelism is orchestrated through a single MeshContext object that holds device meshes, strategy configs, and axis names. <!-- NVSkills catalog signing requested after PR #2937 (2026-07-31). -->

Instructions

For conceptual distributed-training questions, answer directly from the quick patterns in this skill without inspecting the repository. Start with the strategy choice, then list only the YAML fields and constraints relevant to the question.

Use direct action verbs in the final answer: recommend the strategy, show the minimal YAML, state the sizing constraint, and name the unsupported strategies. Do not discuss model onboarding, recipes, Slurm, SkyPilot, or checkpointing unless the user asks.

Examples

TP plus PP for a large multi-node model

Recommend strategy: fsdp2. Mention tpsize, ppsize, cpsize, epsize, and the pipeline sub-config. State that dpsize is inferred from worldsize / (tpsize ppsize cp_size).

distributed:
  strategy: fsdp2
  tp_size: 8
  pp_size: 4
  cp_size: 1
  ep_size: 1
  pipeline:
    pp_schedule: interleaved1f1b
    pp_microbatch_size: 1

MoE expert parallelism

Recommend strategy: fsdp2 with epsize > 1. Say this creates a separate moemesh; include the moe sub-config when relevant; state that epsize must divide dpsize * cpsize. Do not recommend megatronfsdp or ddp.

distributed:
  strategy: fsdp2
  ep_size: 8
  moe:
    reshard_after_forward: false

MegatronFSDP limitations

Say no for pipeline parallelism, expert parallelism, and sequenceparallel. Recommend fsdp2 for PP, EP, or sequenceparallel; mention that DDP is only simple data parallelism.

Strategy Selection

Three strategies are available, selected via the distributed.strategy YAML key:

Decision tree:

• Single GPU: no distributed config needed (FSDP2Manager skips parallelization when world_size=1).
• Multi-GPU single node: fsdp2 (default). Use ddp only if you need the simplest possible setup.
• Multi-node: fsdp2 with appropriate TP/PP sizing.
• MoE models with expert parallelism: fsdp2 with epsize > 1 (creates a separate moemesh).
• Large models (70B+): fsdp2 with PP + TP.
• Long sequences (8K+): add CP (cp_size > 1).

When answering strategy-selection questions, state the chosen distributed.strategy first, then enumerate the YAML fields the user must set.

Quick TP + PP answer:

• Use strategy: fsdp2; do not use megatron_fsdp when pipeline parallelism is required.
• Set tpsize for tensor parallelism and ppsize for pipeline parallelism.
• Add a pipeline: sub-config with ppschedule and ppmicrobatch_size.
• Leave dpsize unset or none; it is inferred as worldsize / (tpsize ppsize cp_size).
• Keep TP inside a fast intra-node domain when possible, and use PP across model depth for 70B+ models.

Quick MoE expert-parallel answer:

• Start with strategy: fsdp2 and ep_size > 1.
• Include a moe: sub-config only when ep_size > 1; it maps to MoEParallelizerConfig.
• Expect a separate moemesh for expert parallelism in addition to the main devicemesh.
• Do not recommend megatronfsdp or ddp for expert parallelism; megatronfsdp has no EP support.
• Before finishing an MoE EP answer, explicitly state that epsize must divide dpsize * cpsize and that megatronfsdp does not support EP, PP, or sequence_parallel.

YAML Config Structure

The distributed section in the recipe YAML maps directly to parsedistributedsection() in recipes/distutils.py:

distributed:
  strategy: fsdp2           # fsdp2 | megatron_fsdp | ddp
  dp_size: none             # auto-calculated from world_size / (tp * pp * cp)
  dp_replicate_size: none   # FSDP2-only, for HSDP
  tp_size: 1
  pp_size: 1
  cp_size: 1
  ep_size: 1

  # Strategy-specific flags (forwarded to the strategy dataclass):
  sequence_parallel: false
  activation_checkpointing: false
  defer_fsdp_grad_sync: true   # FSDP2 only

  # Sub-configs (optional):
  pipeline:
    pp_schedule: 1f1b
    pp_microbatch_size: 1
    # ... see PipelineConfig fields

  moe:
    reshard_after_forward: false
    # ... see MoEParallelizerConfig fields

The dp_size is always inferred:

dp_size = world_size / (tp_size * pp_size * cp_size)

Infrastructure Flow

initialize_distributed()                       [components/distributed/init_utils.py]
    -> initializes torch.distributed process group and returns DistInfo
YAML distributed section + DistInfo.world_size
    -> parse_distributed_section()          [recipes/_dist_utils.py]
    -> create_distributed_setup_from_config()              [recipes/_dist_utils.py]
        -> DistributedSetup.build()         [components/distributed/config.py]
    -> instantiate_infrastructure()         [_transformers/infrastructure.py]
        -> _instantiate_distributed()       -> FSDP2Manager / MegatronFSDPManager / DDPManager
        -> _instantiate_pipeline()          -> AutoPipeline (if pp_size > 1)
        -> parallelize_fn                   -> MoE parallelizer (if ep_size > 1) or PP wrapper
    -> apply_model_infrastructure()         [_transformers/infrastructure.py]
        -> _shard_pp() or _shard_ep_fsdp()  (applies sharding to the model)

FSDP2 Configuration

Basic FSDP2 (data parallelism only)

distributed:
  strategy: fsdp2
  tp_size: 1
  cp_size: 1

This auto-calculates dpsize = worldsize and applies fully_shard() per transformer block via DTensor-based sharding.

FSDP2 with Tensor Parallelism

Keep TP within a single NVLink domain (typically one node):

distributed:
  strategy: fsdp2
  tp_size: 4        # 2, 4, or 8 -- must divide GPUs per node
  sequence_parallel: true

The TP plan is auto-selected based on the model type. Pass a custom plan via the Python API if needed:

config = FSDP2Config(sequence_parallel=True, tp_plan=my_custom_plan)

FSDP2 with Pipeline Parallelism

distributed:
  strategy: fsdp2
  pp_size: 2
  pipeline:
    pp_schedule: interleaved1f1b   # 1f1b, gpipe, interleaved_1f1b, etc.
    pp_microbatch_size: 4
    scale_grads_in_schedule: false

The model must have a ppplan attribute (set on the HF model class) for AutoPipeline to know how to split layers across stages. Models without ppplan are not compatible with PP.

FSDP2 with HSDP (Hybrid Sharded Data Parallel)

Intra-node full sharding + inter-node replication via a 2D DeviceMesh:

distributed:
  strategy: fsdp2
  dp_replicate_size: 2   # must divide dp_size

Constraint: dpreplicatesize < dp_size (pure replication with no sharding is not supported by FSDP2).

Activation Checkpointing

Trades compute for memory by recomputing activations during backward:

distributed:
  activation_checkpointing: true

This is a model-build/training behavior flag, not mesh topology. Dense strategies read it from the strategy config; EP/MoE paths pass the recipe-level flag directly into model infrastructure.

Gradient Sync Deferral

FSDP2 defers gradient sync to the final micro-batch by default for communication overlap:

distributed:
  defer_fsdp_grad_sync: true   # default

Mixed Precision

FSDP2Config defaults to bfloat16 for all three precision knobs via MixedPrecisionPolicy(paramdtype=bf16, reducedtype=bf16, outputdtype=bf16, castforward_inputs=True). Override via the Python API:

from torch.distributed.fsdp import MixedPrecisionPolicy
config = FSDP2Config(
    mp_policy=MixedPrecisionPolicy(param_dtype=torch.float16, reduce_dtype=torch.float32),
)

Pipeline Parallelism

Requirements

1. Model class must define ppplan (a dict mapping module FQNs to stages).
2. pp_size > 1 in the distributed section.
3. A pipeline sub-config with schedule and microbatch size.

Supported schedules

Defined in PipelineConfig.pp_schedule:

• 1f1b (one-forward-one-backward, default)
• gpipe
• interleaved_1f1b / interleaved1f1b
• looped_bfs
• dfs
• v_schedule
• zero_bubble

Example (8B model on 8 GPUs, PP=2 + DP=4)

distributed:
  strategy: fsdp2
  pp_size: 2

  pipeline:
    pp_schedule: interleaved1f1b
    pp_microbatch_size: 4
    scale_grads_in_schedule: false

checkpoint:
  model_save_format: safetensors
  save_consolidated: final

How it works

AutoPipeline.build() calls pipelinemodel() which splits the model into stages using the model's pp_plan, creates PipelineStage objects, and builds the schedule. During training, schedule.step() drives forward and backward through the pipeline.

Context Parallelism

Use CP for long sequences (8K+). CP shards Q/K/V on the sequence dimension as DTensors.

Config

distributed:
  strategy: fsdp2
  cp_size: 2   # or 4, 8

Requirements

• SDPA (Flash Attention or Efficient Attention backend) or Transformer Engine

attention. SDPBackend.MATH is not compatible with DTensor.

• Attention masks are automatically stripped; is_causal=True is set via

forward pre-hooks registered by attachcontextparallel_hooks().

How it works

1. After model sharding, applymodelinfrastructure() calls

attachcontextparallel_hooks() on each model part (for non-TE models).

1. At each training step, makecpbatchandctx() creates a CP context

manager that shards the batch along the sequence dimension and sets up context_parallel() from torch.distributed.tensor.experimental.

1. For TE attention models, makecpbatchforte() uses THD format and

TE's thdgetpartitioned_indices for sharding.

CP with Sequence Packing

CP works with packed sequences. The packedsequencesize must be divisible by cpsize. When using TE, chunks are sharded per-chunk via shardthdchunkforte().

Sequence Packing

Packing multiple sequences into a single training sample for efficiency.

Config

packed_sequence:
  packed_sequence_size: 4096   # 0 = disabled

step_scheduler:
  local_batch_size: 1          # must be 1 for packed sequences

When packedsequencesize > 0, the dataset collator packs sequences up to that length. localbatchsize must be 1 because each "sample" is already a packed batch.

MoE Distributed Training

Expert Parallelism

Set epsize > 1 to distribute experts across GPUs. This creates a separate moemesh alongside the main device_mesh:

distributed:
  strategy: fsdp2
  ep_size: 8
  activation_checkpointing: true

The moemesh shape is (ppsize, epshardsize, epsize) with dimension names ("pp", "epshard", "ep").

Constraint: dpcpsize (= dpsize * cpsize) must be divisible by ep_size.

MoE sub-config

distributed:
  strategy: fsdp2
  ep_size: 8
  activation_checkpointing: true

  moe:
    reshard_after_forward: false
    ignore_router_for_ac: false
    wrap_outer_model: true

The moe sub-section maps to MoEParallelizerConfig and is only instantiated when ep_size > 1.

Full MoE example (Qwen3-30B-A3B on 8 GPUs)

distributed:
  strategy: fsdp2
  tp_size: 1
  cp_size: 1
  pp_size: 1
  ep_size: 8
  sequence_parallel: false
  activation_checkpointing: true

MegatronFSDP limitations

Despite its name, megatronfsdp does not support expert parallelism (epsize > 1), pipeline parallelism (ppsize > 1), or sequenceparallel. Use fsdp2 for these features.

Parallelism Sizing Guidelines

Dense models

MoE models

MoE models need less TP than dense models of similar total parameter count because only a fraction of parameters are active per token. EP is the primary scaling dimension:

Hardware topology rules

• TP must stay within a single NVLink domain (one node, typically 8 GPUs).
• Use PP or DP for cross-node scaling.
• TP across InfiniBand degrades throughput severely.

Programmatic API (frompretrained / fromconfig)

When not using YAML recipes, configure distributed training via Python:

from nemo_automodel.components.distributed import (
    DistributedSetup,
    FSDP2Config,
    ParallelismSizes,
    initialize_distributed,
)

dist_env = initialize_distributed("nccl")
distributed_setup = DistributedSetup.build(
    strategy=FSDP2Config(sequence_parallel=True),
    parallelism_sizes=ParallelismSizes(tp_size=2),
    activation_checkpointing=True,
    world_size=dist_env.world_size,
)

Or pass directly to from_pretrained:

from nemo_automodel import NeMoAutoModelForCausalLM

model = NeMoAutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.2-1B",
    distributed_setup=distributed_setup,
)

Code Anchors

Strategy config dataclasses:

components/distributed/config.py
    FSDP2Config       -- sequence_parallel, tp_plan, mp_policy, offload_policy,
                         activation_checkpointing, defer_fsdp_grad_sync
    MegatronFSDPConfig -- zero_dp_strategy, overlap_grad_reduce, overlap_param_gather, etc.
    DDPConfig          -- activation_checkpointing only

MeshContext (single source of truth for parallelism):

components/distributed/mesh.py
    MeshContext  -- device_mesh, moe_mesh
                    Properties: tp_size, pp_size, cp_size, ep_size, dp_size, dp_replicate_size
    MeshAxisName -- PP, DP, DP_REPLICATE, DP_SHARD, DP_SHARD_CP, DP_CP, CP, TP, EP, EP_SHARD

Mesh context and raw mesh creation:

components/distributed/config.py
    DistributedSetup.build()      -- builds MeshContext from strategy + parallelism
components/distributed/mesh_utils.py
    _create_device_meshes()       -- routes to FSDP2/MegatronFSDP/DDP raw mesh creation
    _create_fsdp2_device_mesh()   -- shape (pp, dp_replicate, dp_shard, cp, tp) + flattened submeshes
    _create_megatron_fsdp_device_mesh() -- shape (dp, cp, tp)

Distributed managers:

components/distributed/fsdp2.py          -- FSDP2Manager.parallelize()
components/distributed/megatron_fsdp.py  -- MegatronFSDPManager.parallelize()
components/distributed/ddp.py            -- DDPManager

Pipeline parallelism:

components/distributed/pipelining/config.py        -- PipelineConfig dataclass
components/distributed/pipelining/autopipeline.py  -- AutoPipeline orchestrator
components/distributed/pipelining/functional.py    -- pipeline_model(), schedule creation
components/distributed/pipelining/hf_utils.py      -- HF model validation for PP

Context parallelism:

components/distributed/context_parallel/utils.py
    make_cp_batch_and_ctx()            -- creates CP context manager + shards batch
    create_context_parallel_ctx()      -- wraps torch.distributed.tensor.experimental.context_parallel
    attach_context_parallel_hooks()    -- strips attention_mask, sets is_causal=True
    make_cp_batch_for_te()             -- TE-specific CP batch sharding (THD format)

Infrastructure orchestration:

_transformers/infrastructure.py
    instantiate_infrastructure()    -- config objects -> runtime objects
    apply_model_infrastructure()    -- applies sharding, PEFT, checkpoints to model
    _shard_pp()                     -- pipeline parallel path
    _shard_ep_fsdp()                -- EP + FSDP path (non-PP)

YAML parsing:

recipes/_dist_utils.py
    parse_distributed_section()  -- YAML dict -> typed configs + sizes
    create_distributed_setup_from_config()  -- recipe adapter: parse + create DistributedSetup; does not init process group

MoE config:

components/distributed/config.py
    MoEParallelizerConfig  -- reshard_after_forward, ignore_router_for_ac, wrap_outer_model, etc.
components/moe/config.py
    MoEConfig              -- n_routed_experts, n_activated_experts, score_func, etc.

Pitfalls

1. TP across nodes destroys throughput. Always keep TP within a single

NVLink domain. Use PP or DP for cross-node scaling.

1. PP requires ppplan on the model class. Not all HF models have this.

Check validatehfmodelforpipeline_support() before enabling PP.

1. PP bubbles reduce GPU utilization. Use interleaved schedules

(interleaved_1f1b) and smaller microbatches to reduce bubble time.

1. FSDP2 requires DTensor-aware state dict saving. Use safetensors with

saveconsolidated: final for final HF export, or saveconsolidated: false plus the generated model/consolidate.sh helper for offline export.

1. CP requires compatible attention. SDPA (Flash Attention or Efficient

Attention) or TE attention only. SDPBackend.MATH is not compatible with DTensor.

1. **MoE EP size must evenly divide dpsize * cpsize.** The device mesh

creation asserts dpcpsize % ep_size == 0.

1. MegatronFSDP is more limited than FSDP2. It does not support PP

(ppsize > 1), EP (epsize > 1), or sequence_parallel. The MeshContext validation raises on these combinations.

1. DDP supports nothing beyond data parallelism. No TP, PP, CP, EP, or

HSDP. Validation raises on any of these.

1. Activation checkpointing increases compute. It saves memory by

recomputing activations during backward, but adds ~30% compute overhead.

1. Mixed precision policy must match model expectations. The default

bfloat16 policy works for most models. FP16 models may need a custom MixedPrecisionPolicy.

1. packedsequencesize must be divisible by cp_size when using CP

with packed sequences.

1. dpreplicatesize is FSDP2-only. Passing it with megatron_fsdp

or ddp raises a ValueError.

Verification

Run the smallest recipe that exercises the requested strategy. Success means exit code 0, finite loss, no NCCL timeout, and log output matching the expected TP/PP/CP/EP sizes.