diff --git a/docs/assets/mesh_spmd2.png b/docs/assets/mesh_spmd2.png new file mode 100644 index 00000000000..cd7bf793711 Binary files /dev/null and b/docs/assets/mesh_spmd2.png differ diff --git a/docs/assets/spmd_mode.png b/docs/assets/spmd_mode.png new file mode 100644 index 00000000000..dd9b5cc69cc Binary files /dev/null and b/docs/assets/spmd_mode.png differ diff --git a/docs/source/index.rst b/docs/source/index.rst index 403946e892a..ed254416af7 100644 --- a/docs/source/index.rst +++ b/docs/source/index.rst @@ -85,3 +85,4 @@ test .. mdinclude:: ../fsdp.md .. mdinclude:: ../ddp.md .. mdinclude:: ../gpu.md +.. mdinclude:: ../spmd.md diff --git a/docs/spmd.md b/docs/spmd.md new file mode 100644 index 00000000000..69dd1f14853 --- /dev/null +++ b/docs/spmd.md @@ -0,0 +1,338 @@ + +# PyTorch/XLA SPMD User Guide + +In this user guide, we discuss how [GSPMD](https://arxiv.org/abs/2105.04663) is integrated in PyTorch/XLA, and provide a design overview to illustrate how the SPMD sharding annotation API and its constructs work. And then, we provide a list of reference examples for users to try. + + +## What is PyTorch/XLA SPMD? + +[GSPMD](https://arxiv.org/abs/2105.04663) is an automatic parallelization system for common ML workloads. The XLA compiler will transform the single device program into a partitioned one with proper collectives, based on the user provided sharding hints. This feature allows developers to write PyTorch programs as if they are on a single large device without any custom sharded computation ops and/or collective communications to scale. + +![alt_text](assets/spmd_mode.png "image_tooltip") +_Figure 1. Comparison of two different execution strategies, (a) for non-SPMD and (b) for SPMD._ + +To support GSPMD in PyTorch/XLA, we are introducing a new execution mode. Before GSPMD, the execution mode in PyTorch/XLA assumed multiple model replicas, each with a single core (Figure 1.a). This mode of execution, as illustrated in the above suits data parallelism frameworks, like the popular PyTorch [Distributed Data Parallel (DDP)](https://pytorch.org/tutorials/intermediate/ddp_tutorial.html) or Fully Sharded Data Parallel (FSDP), but is also limited in that a replica can only reside on one device core for execution. PyTorch/XLA SPMD introduces a new execution mode that assumes a single replica with multiple cores (Figure 1.b), allowing a replica to run across multiple device cores. This shift unlocks more advanced parallelism strategies for better large model training performance. + +PyTorch/XLA SPMD is available on the new [PJRT](https://github.com/pytorch/xla/blob/master/docs/pjrt.md) runtime. To enable PyTorch/XLA SPMD execution mode, the user must call [use_spmd() API](https://github.com/pytorch/xla/blob/b8b484515a97f74e013dcf38125c44d53a41f011/torch_xla/runtime.py#L214). + +```python +import torch_xla.runtime as xr + +# Enable PyTorch/XLA SPMD execution mode. +xr.use_spmd() +assert xr.is_spmd() == True +``` + +It is important to note that SPMD is a replacement for any existing parallel mechanisms, including DDP and FSDP. Users can not mix two different execution modes (SPMD and non-SPMD), and later in this guide we will go over how to use SPMD annotation to perform DDP and FSDP. + +Also, this version of the SPMD is currently only tested.optimized on Google Cloud TPU. GPU support and optimization will come in the 2.2 release. + + +## PyTorch/XLA SPMD Design Overview + + +### Simple Eexample & Sharding Aannotation API + +Users can annotate native PyTorch tensors using the `mark_sharding` API ([src](https://github.com/pytorch/xla/blob/9a5fdf3920c18275cf7dba785193636f1b39ced9/torch_xla/experimental/xla_sharding.py#L388)). This takes `torch.Tensor` as input and returns a `XLAShardedTensor` as output. + +```python +def mark_sharding(t: Union[torch.Tensor, XLAShardedTensor], mesh: Mesh, partition_spec: Tuple[Union[int, None]]) -> XLAShardedTensor +``` + +Invoking `mark_sharding` API takes a user defined logical [mesh](#mesh) and [partition\_spec](#partition-spec) and generates a sharding annotation for the XLA compiler. The sharding spec is attached to the XLATensor. Here is a simple usage example from the [[RFC](https://github.com/pytorch/xla/issues/3871), to illustrate how the sharding annotation API works: + +```python +import torch +import torch_xla.core.xla_model as xm +import torch_xla.runtime as xr +import torch_xla.experimental.xla_sharding as xs +from torch_xla.experimental.xla_sharding import Mesh + +# Enable XLA SPMD execution mode. +xr.use_spmd() + +# Device mesh, this and partition spec as well as the input tensor shape define the individual shard shape. +mesh_shape = (2, 4) +num_devices = xr.global_runtime_device_count() +device_ids = np.array(range(num_devices)) +mesh = Mesh(device_ids, mesh_shape, ('x', 'y')) + +t = torch.randn(8, 4).to(xm.xla_device()) + +# Mesh partitioning, each device holds 1/8-th of the input +partition_spec = (0, 1) +m1_sharded = xs.mark_sharding(t, mesh, partition_spec) +assert isinstance(m1_sharded, XLAShardedTensor) == True +``` + +We can annotate different tensors in the PyTorch program to enable different parallelism techniques, as described in the comment below: + +```python +# Sharding annotate the linear layer weights. +model = SimpleLinear().to(xm.xla_device()) +xs.mark_sharding(model.fc1.weight, mesh, partition_spec) + +# Training loop +model.train() +for step, (data, target) in enumerate(loader): + # Assumes `loader` returns data, target on XLA device + optimizer.zero_grad() + # Sharding annotate input data, we can shard any input + # dimensions. Sharidng the batch dimension enables + # in data parallelism, sharding the feature dimension enables + # spatial partitioning. + xs.mark_sharding(data, mesh, partition_spec) + ouput = model(data) + loss = loss_fn(output, target) + optimizer.step() + xm.mark_step() +``` + +More complete unit test cases and integration test examples are available in the PyTorch/XLA [repo](https://github.com/pytorch/xla/tree/r2.0/test/spmd). + + +### Mesh + +For a given cluster of devices, a physical mesh is a representation of the interconnect topology. + +We derive a logical mesh based on this topology to create sub-groups of devices which can be used for partitioning different axes of tensors in a model. + +![alt_text](assets/mesh_spmd2.png "image_tooltip") + +We abstract logical mesh with [Mesh API](https://github.com/pytorch/xla/blob/028df4da388468fa9a41b1f98ea08bfce13b4c63/torch_xla/experimental/xla_sharding.py#L16). The axes of the logical Mesh can be named. Here is an example: + +```python +import torch_xla.runtime as xr +from torch_xla.experimental.xla_sharding import Mesh + +# Assuming you are running on a TPU host that has 8 devices attached +num_devices = xr.global_runtime_device_count() +# mesh shape will be (4,2) in this example +mesh_shape = (num_devices // 2, 2) +device_ids = np.array(range(num_devices)) +# axis_names 'x' nad 'y' are optional +mesh = Mesh(device_ids, mesh_shape, ('x', 'y')) + +mesh.get_logical_mesh() +>> array([[0, 1], + [2, 3], + [4, 5], + [6, 7]]) +mesh.shape() +>> OrderedDict([('x', 4), ('y', 2)]) +``` + +In general, SPMD programs should create a single mesh and reuse it for all sharding to ensure that the tiling assignment is consistent with the intended sharding strategy. The same mesh can be reused for tensors of different shapes and shardings by manipulating the partition spec, described further below. + +### Hybrid Mesh + +Mesh nicely abstracts how the physical device mesh is constructed. Users can arrange devices in any shape and order using the logical mesh. However, one can define a more performant mesh based on the physical topology, especially when it involves Data Center Network (DCN) cross slice connections. HybridMesh creates a mesh which gives good performance out of the box for such multislice environments. It accepts ici\_mesh\_shape and dcn\_mesh\_shape which denote logical mesh shapes of inner and outer network. + +```python +from torch_xla.experimental.xla_sharding import HybridMesh + +# This example is assuming 2 slices of v4-8. +# - ici_mesh_shape: shape of the logical mesh for inner connected devices. +# - dcn_mesh_shape: shape of logical mesh for outer connected devices. +ici_mesh_shape = (1, 4, 1) # (data, fsdp, tensor) +dcn_mesh_shape = (2, 1, 1) + +mesh = HybridMesh(ici_mesh_shape, dcn_mesh_shape, ('data','fsdp','tensor')) +print(mesh.shape()) +>> OrderedDict([('data', 2), ('fsdp', 4), ('tensor', 1)]) +``` + + +### Partition Spec + +partition\_spec has the same rank as the input tensor. Each dimension describes how the corresponding input tensor dimension is sharded across the device mesh (logically defined by mesh\_shape). `partition_spec` is a tuple of `device_mesh` dimension `index` or None. The index can be an `int` or `str`, if the corresponding mesh dimension is named. This specifies how each input rank is sharded (`index` to `mesh_shape`) or replicated (`None`). + +```python +# Provide optional mesh axis names and use them in the partition spec +mesh = Mesh(device_ids, (4, 2), ('data', 'model')) +partition_spec = ('model', 'data') +xs.mark_sharding(input_tensor, mesh, partition_spec) +``` + +We support all three types of sharding, described in the original [GSPMD](https://arxiv.org/abs/2105.04663) paper. For instance, one can specify partial replication like this: + +```python +# Provide optional mesh axis names and use them in the partition spec +mesh = Mesh(device_ids, (2, 2, 2), ('x', 'y', 'z')) + +# evenly shard across x and z and replicate among y +partition_spec = ('x', 'z') # equivalent to ('x', None, 'z') +xs.mark_sharding(input_tensor, mesh, partition_spec) +``` + +The partition spec enables reuse of the same mesh for different tensor shapes and desired sharding strategies. The following example demonstrates this using a 3D mesh: + +```python +# Create a 3-D mesh of 8 devices with logical dimensions replica, fsdp, and +# tensor +mesh = Mesh(device_ids, (2, 2, 2), ('replica', 'fsdp', 'tensor')) + +# A 2D tensor can be sharded along the fsdp and tensor axes and replicated +# along the replica axis by omitting `replica` from the partition spec. +two_d_partially_replicated = torch.randn(64, 64, device='xla') +xs.mark_sharding(two_d_partially_replicated, mesh, ('fsdp', 'tensor')) + +# A 2D tensor can be sharded across all dimensions by combining, for example, +# the replica and fsdp mesh axes using a tuple +two_d_fully_sharded = torch.randn(64, 64, device='xla') +xs.mark_sharding(two_d_fully_sharded, mesh, (('replica', 'fsdp'), 'tensor')) + +# A 4D tensor can be sharded along up to three of its axes using the 3D mesh +four_d = torch.randn(64, 64, 64, 64, device='xla') +xs.mark_sharding(four_d, ('replica', 'fsdp', None, 'tensor')) +``` + + +### XLAShardedTensor + +The main use case for `XLAShardedTensor` [[RFC](https://github.com/pytorch/xla/issues/3871)] is to annotate a native `torch.tensor` (on a single device) with a sharding spec. The annotation takes place immediately, but the actual sharding of the tensor is delayed as the computation is carried out lazily, except for the input tensors which are sharded without delay. Once a tensor is annotated and wrapped inside a `XLAShardedTensor`, it can be passed to existing PyTorch ops and `nn.Module` layers as `torch.Tensor`. This is important to ensure that the same PyTorch layers and tensor ops can be stacked together with `XLAShardedTensor`. This means that the user does not need to rewrite the existing ops and model codes for sharded computation. Namely, `XLAShardedTensor` will satisfy the following requirements: + + + +* `XLAShardedTensor` is a `torch.Tensor` subclass and works directly with native torch ops and `module.layers`. We use `__torch_dispatch__` to send `XLAShardedTensor` to the XLA backend. PyTorch/XLA retrieves attached sharding annotations to trace the graph and invokes XLA SPMDPartitioner. +* Internally, `XLAShardedTensor` (and its global\_tensor input) is backed by `XLATensor` with a special data structure holding references to the sharded device data. +* The sharded tensor after lazy execution may be gathered and materialized back to the host as global\_tensor when requested on the host (e.g., printing the value of the global tensor. +* The handles to the local shards are materialized strictly after the lazy execution. `XLAShardedTensor` exposes [local\_shards](https://github.com/pytorch/xla/blob/909f28fa4c1a44efcd21051557b3bcf2d399620d/torch_xla/experimental/xla_sharded_tensor.py#L111) to return the local shards on addressable devices as List[[XLAShard](https://github.com/pytorch/xla/blob/909f28fa4c1a44efcd21051557b3bcf2d399620d/torch_xla/experimental/xla_sharded_tensor.py#L12)]. + +There is also an ongoing effort to integrate XLAShardedTensor into DistributedTensor API to support XLA backend [[RFC](https://github.com/pytorch/pytorch/issues/92909)]. + + +### Sharding-Aware Host-to-Device Data Loading + +PyTorch/XLA SPMD takes a single-device program, shards and executes it in parallel. The SPMD execution requires using the native PyTorch DataLoader, which transfers data synchronously from the host to XLA devices. This blocks the training during the input data transfer every step. To improve the native data loading performance, we made PyTorch/XLA ParallelLoader support input sharding directly (src), when passed the optional kwarg _input\_sharding_: + +```python +# MpDeviceLoader returns ParallelLoader.per_device_loader as iterator +train_loader = pl.MpDeviceLoader( + train_loader, # wraps PyTorch DataLoader + device, + # optional input_sharding field + input_sharding=xs.ShardingSpec(input_mesh, (0, 1, 2, 3))) +``` + + +### Distributed Checkpointing + +PyTorch/XLA SPMD is compatible with the [torch.distributed.checkpoint](https://pytorch.org/docs/stable/distributed.checkpoint.html) library through a dedicated `Planner` instance. Users are able to synchronously save and load checkpoints through this common interface. + +The SPMDSavePlanner and SPMDLoadPlanner ([src](https://github.com/pytorch/xla/blob/master/torch_xla/experimental/distributed_checkpoint.py)) classes enable the `save_state_dict` and `load_state_dict` functions to operate directly on the shards of an `XLAShardedTensor`, enabling all of the benefits of distributed checkpointing in SPMD training. + +Here is a demonstration of the synchronous distributed checkpointing API: + +```python +import torch.distributed.checkpoint as dist_cp +import torch_xla.experimental.distributed_checkpoint as xc + +# Saving a state_dict +state_dict = { + "model": model.state_dict(), + "optim": optim.state_dict(), +} + +dist_cp.save_state_dict( + state_dict=state_dict, + storage_writer=dist_cp.FileSystemWriter(CHECKPOINT_DIR), + planner=xc.SPMDSavePlanner(), +) +... + +# Loading the model's state_dict from the checkpoint. The model should +# already be on the XLA device and have the desired sharding applied. +state_dict = { + "model": model.state_dict(), +} + +dist_cp.load_state_dict( + state_dict=state_dict, + storage_reader=dist_cp.FileSystemReader(CHECKPOINT_DIR), + planner=xc.SPMDLoadPlanner(), +) +model.load_state_dict(state_dict["model"]) +``` + +### Virtual Device Optimization + +PyTorch/XLA normally transfers tensor data asynchronously from host to device once the tensor is defined. This is to overlap the data transfer with the graph tracing time. However, because GSPMD allows the user to modify the tensor sharding _after _the tensor has been defined, we need an optimization to prevent unnecessary transfer of tensor data back and forth between host and device. We introduce Virtual Device Optimization, a technique to place the tensor data on a virtual device SPMD:0 first, before uploading to the physical devices when all the sharding decisions are finalized. Every tensor data in SPMD mode is placed on a virtual device, SPMD:0. The virtual device is exposed to the user as an XLA device XLA:0 with the actual shards on physical devices, like TPU:0, TPU:1, etc. + + +### Number of processes + +Unlike existing DDP and FSDP, under the SPMD mode, there is always a single process running on each accelerator host. This provides the benefit that PyTorch/XLA only need to compile each graph once which can be reused for all accelerators attached to this host. + + +### Running SPMD on TPU Pod + +There is no code change required to go from single TPU host to TPU Pod if you construct your mesh and partition spec based on the number of devices instead of some hardcode constant. To run the PyTorch/XLA workload on TPU Pod, please refer to the [Pods section](https://github.com/pytorch/xla/blob/master/docs/pjrt.md#pods) of our PJRT guide. + + +## Reference Examples + + +### Use SPMD to express Data Parallel + +The SPMD API is general enough to express both data parallelism and model parallelism. One can implement data parallelism simply by annotating the input batch dimension for sharding. Here, we have shard the batch dimension across all available devices (N-way):There are 2 ways of using SPMD to express data parallel or batch sharding: + +```python +num_devices = xr.global_runtime_device_count() + +# Assume data is 4d and 0th dimension is the batch dimension +mesh_shape = (num_devices, 1, 1, 1) +input_mesh = xs.Mesh(device_ids, mesh_shape, ('B', 'C', 'W', 'H')) +partition_spec = range(num_devices) + +# Shard the batch dimension +xs.mark_sharding(input_tensor, input_mesh, partition_spec) +``` + +PyTorch/XLA’s MpDeviceLoader supports input batch sharding, which also loads the batches to the devices in the background: + +```python +num_devices = xr.global_runtime_device_count() + +# Assume data is 4d and 0th dimension is the batch dimension +mesh_shape = (num_devices, 1, 1, 1) +input_mesh = xs.Mesh(device_ids, mesh_shape, ('B', 'C', 'W', 'H')) +partition_spec = range(num_devices) + +# Use MpDeviceLoader to load data in background +train_loader = pl.MpDeviceLoader( + train_loader, + device, + input_sharding=xs.ShardingSpec(input_mesh, partition_spec)) +``` + +We highly recommend the second approach as it should yield a better training performance. + + +### Use SPMD to express FSDP(Fully Sharded Data Parallel) + +PyTorch’s FSDP is data parallel + sharded model parameters at 0th dimension. Users first need to use SPMD to express Data Parallels as suggested in the previous section. + +```python +for name, param in model.named_parameters(): + shape = (num_devices,) + (1,) * (len(param.shape) - 1) + mesh = xs.Mesh(device_ids, shape) + xs.mark_sharding(param, mesh, range(len(param.shape))) +``` + + +### Running Resnet50 example with SPMD + +We provided a quick example of [resnet50](https://github.com/pytorch/xla/blob/master/test/spmd/test_train_spmd_imagenet.py) with a couple different SPMD sharding strategies for you to play around with. You can first run it without SPMD using + +```bash +python test/spmd/test_train_spmd_imagenet.py --fake_data --batch_size 512 +``` + +and check the throughput. After that you can enable the batch sharding with + +```bash +XLA_USE_SPMD=1 python test/spmd/test_train_spmd_imagenet.py --fake_data --batch_size 2048 --model=resnet50 --sharding=batch +``` + +Note that I used a batch size 4 times as large since I am running it on a TPU v4 which has 4 TPU devices attached to it. You should see the throughput becomes roughly 4x the non-spmd run.