[Feature]: support FP8 communication in DDP, FSDP, Gemini (#5928)

* support fp8_communication in the Torch DDP grad comm, FSDP grad comm, and FSDP params comm

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* implement communication hook for FSDP params all-gather

* added unit test for fp8 operators

* support fp8 communication in GeminiPlugin

* update training scripts to support fsdp and fp8 communication

* fixed some minor bugs observed in unit test

* add all_gather_into_tensor_flat_fp8

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* add skip the test if torch < 2.2.0

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* add skip the test if torch < 2.2.0

* add skip the test if torch < 2.2.0

* add fp8_comm flag

* rebase latest fp8 operators

* rebase latest fp8 operators

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

colossalai/booster/plugin/gemini_plugin.py

@@ -364,6 +364,7 @@ class GeminiPlugin(DPPluginBase):
         enable_sequence_overlap: bool = False,
         enable_async_reduce: bool = True,
         verbose: bool = False,
+        fp8_communication: bool = False,
     ) -> None:
         super().__init__()
         assert precision in SUPPORTED_PRECISION, f"precision {precision} is not supported"
@@ -395,6 +396,7 @@ class GeminiPlugin(DPPluginBase):
             master_weights=master_weights,
             max_prefetch=max_prefetch,
             enable_async_reduce=enable_async_reduce,
+            fp8_communication=fp8_communication,
         )
         self.zero_optim_config = dict(
             gpu_margin_mem_ratio=gpu_margin_mem_ratio,

colossalai/booster/plugin/torch_ddp_plugin.py

@@ -177,6 +177,7 @@ class TorchDDPPlugin(DPPluginBase):
         check_reduction: bool = False,
         gradient_as_bucket_view: bool = False,
         static_graph: bool = False,
+        fp8_communication: bool = False,
     ) -> None:
         super().__init__()
         self.ddp_kwargs = dict(
@@ -187,6 +188,7 @@ class TorchDDPPlugin(DPPluginBase):
             gradient_as_bucket_view=gradient_as_bucket_view,
             static_graph=static_graph,
         )
+        self.fp8_communication = fp8_communication
 
     def support_no_sync(self) -> bool:
         return True
@@ -226,6 +228,11 @@ class TorchDDPPlugin(DPPluginBase):
         if optimizer is not None and not isinstance(optimizer, OptimizerWrapper):
             optimizer = OptimizerWrapper(optimizer)
 
+        if self.fp8_communication:
+            from colossalai.quantization.fp8 import fp8_compress_ddp_grad_comm_hook_async
+
+            model.module.register_comm_hook(None, fp8_compress_ddp_grad_comm_hook_async)
+
         return model, optimizer, criterion, dataloader, lr_scheduler
 
     def control_checkpoint_io(self) -> bool:

colossalai/booster/plugin/torch_fsdp_plugin.py

@@ -298,6 +298,7 @@ class TorchFSDPPlugin(DPPluginBase):
             ignored_modules: Optional[Iterable[torch.nn.Module]] = None,
             param_init_fn: Optional[Callable[[nn.Module], None]] = None,
             sync_module_states: bool = False,
+            fp8_communication: bool = False,
         ):
             super().__init__()
             self.fsdp_kwargs = dict(
@@ -311,6 +312,7 @@ class TorchFSDPPlugin(DPPluginBase):
                 param_init_fn=param_init_fn,
                 sync_module_states=sync_module_states,
             )
+            self.fp8_communication = fp8_communication
 
     else:
         raise RuntimeError("FSDP is not supported while torch version under 1.12.0.")
@@ -347,6 +349,19 @@ class TorchFSDPPlugin(DPPluginBase):
         # wrap the model with PyTorch FSDP
         fsdp_model = TorchFSDPModel(model, device_id=torch.cuda.current_device(), **self.fsdp_kwargs)
 
+        if self.fp8_communication:
+            from colossalai.quantization.utils import patch_fsdp_params_comm_hook
+
+            patch_fsdp_params_comm_hook()
+
+            from colossalai.quantization.fp8 import fp8_compress_fsdp_params_comm_hook
+
+            fsdp_model.module.register_params_comm_hook(None, fp8_compress_fsdp_params_comm_hook)
+
+            from colossalai.quantization.fp8 import fp8_compress_fsdp_grad_comm_hook
+
+            fsdp_model.module.register_comm_hook(None, fp8_compress_fsdp_grad_comm_hook)
+
         if optimizer is not None:
             if len(optimizer.param_groups) > 1:
                 warnings.warn(

colossalai/quantization/fp8.py

@@ -15,6 +15,7 @@ def cast_to_fp8(inp: torch.Tensor, fp8_format="e4m3", per_channel_scale=False) -
         scale: scaling factor for fp8 casting. If it is None, then it is computed automatically. Per-channel scaling
         is applied if input tensor is 2 dimension, otherwise, per-tensor scaling is applied.
         fp8_format: e4m3 or e5m2
+
     Returns:
         Tuples: A tuple (fp8_tensor, scale)
     """
@@ -29,12 +30,13 @@ def cast_to_fp8(inp: torch.Tensor, fp8_format="e4m3", per_channel_scale=False) -
         per_channel_max = inp.abs().max(dim=-1).values.float()
         per_channel_max = torch.where(per_channel_max > 0, per_channel_max, 1.0)
         scale = fp8_max / per_channel_max[:, None]
+        scale_inv = per_channel_max / fp8_max
     else:
         per_tensor_max = inp.abs().max().float()
         per_tensor_max = torch.where(per_tensor_max > 0, per_tensor_max, 1.0)
         scale = fp8_max / per_tensor_max
-
         scale_inv = 1.0 / scale
+
     ret = (scale * inp.float()).to(fp8_type)
     return ret, scale_inv
 
@@ -185,7 +187,11 @@ def cast_to_fp8_pipeline(inp: Any) -> None:
         return
 
     assert "hidden_states" in inp, "required by pipeline parallelism."
+    assert (
+        inp["hidden_states"].size(-1) % 2 == 0
+    ), "tensor size(-1) must be divisible by 2 to view Float8_e4m3fn as BFloat16 or Float16"
     inp_tensor = inp["hidden_states"]
+    inp_dtype = inp_tensor.dtype
 
     min_val, max_val = inp_tensor.aminmax()
     amax = torch.maximum(min_val.abs(), max_val.abs())
@@ -206,6 +212,7 @@ def cast_to_fp8_pipeline(inp: Any) -> None:
     inp_tensor.data = q_tensor.to(fp8_type).view(fp8_view_type)
 
     inp["fp8_scale"] = scale.float().reciprocal()
+    inp["dtype"] = torch.zeros_like(scale).to(inp_dtype)
 
 
 def cast_from_fp8_pipeline(inp: Any, del_metadata=True) -> None:
@@ -230,10 +237,11 @@ def cast_from_fp8_pipeline(inp: Any, del_metadata=True) -> None:
     else:
         raise TypeError("Only float16, bfloat16 are implemented.")
 
-    inp_tensor.data = inp_tensor.data.view(fp8_type).to(torch.float16) * scale
+    inp_tensor.data = inp_tensor.data.view(fp8_type).to(inp["dtype"]) * scale
 
     if del_metadata:
         del inp["fp8_scale"]
+        del inp["dtype"]
 
 
 def reduce_scatter_fp8(output: torch.Tensor, input_list, group, fp8_format="e5m2") -> None:
@@ -273,6 +281,199 @@ def reduce_scatter_fp8(output: torch.Tensor, input_list, group, fp8_format="e5m2
     output.data = summed_out
 
 
+def fp8_compress_ddp_grad_comm_hook_async(
+    process_group: dist.ProcessGroup,
+    bucket: dist.GradBucket,
+    fp8_format: str = "e5m2",
+) -> torch.futures.Future[torch.Tensor]:
+    """
+    Compress by casting ``GradBucket`` to FP8 floating-point format divided by process group size.
+
+    This DDP communication hook implements a simple gradient compression approach that casts ``GradBucket`` tensor
+    to FP8 floating-point format (``torch.float8_e5m2`` or ``torch.bfloat16_e4m3``), and then divides it
+    by the process group size.
+    Once compressed gradient tensors are allreduced, the chained callback ``decompress`` casts it back
+    to the input data type (such as ``float32``).
+
+    Example::
+        >>> ddp_model.register_comm_hook(process_group, fp8_compress_ddp_grad_comm_hook_async)
+    """
+    group_to_use = process_group if process_group is not None else dist.group.WORLD
+
+    input_tensor = bucket.buffer()
+    world_size = dist.get_world_size()
+    input_type = input_tensor.dtype
+    input_device = input_tensor.device
+    flat_padded_x = input_tensor.flatten()
+
+    if flat_padded_x.size(0) % world_size != 0:
+        pad_size = world_size - flat_padded_x.size(0) % world_size
+        flat_padded_x = F.pad(flat_padded_x, (0, pad_size))
+
+    fp8_type = torch.float8_e4m3fn if fp8_format == "e4m3" else torch.float8_e5m2
+    ret, scale = cast_to_fp8(flat_padded_x, fp8_format=fp8_format)
+
+    inp = ret.view(torch.uint8)
+    output_chunks_single = torch.empty_like(inp)
+    split_sizes = [inp.numel() // world_size for _ in range(world_size)]
+    fut0 = dist.all_to_all_single(
+        output_chunks_single,
+        inp,
+        output_split_sizes=split_sizes,
+        input_split_sizes=split_sizes,
+        group=group_to_use,
+        async_op=True,
+    ).get_future()
+
+    scale_list = [torch.ones(1, dtype=scale.dtype, device=input_device) for _ in range(world_size)]
+    fut1 = dist.all_gather_into_tensor(
+        torch.cat(scale_list, dim=0), scale, group=group_to_use, async_op=True
+    ).get_future()
+    all_to_all_fut = torch.futures.collect_all([fut0, fut1])
+
+    def sum_and_allgather(fut):
+        output_chunks_single = fut.value()[0].wait()[0]
+        scale_list_single = fut.value()[1].wait()[0]
+
+        output_chunks = list(torch.chunk(output_chunks_single, world_size, dim=0))
+        scale_list = scale_list_single.chunk(world_size, dim=0)
+
+        summed_out = torch.zeros_like(output_chunks[0]).to(input_type)
+        for scale, out in zip(scale_list, output_chunks):
+            out = out.view(fp8_type)
+            summed_out += cast_from_fp8(out, scale, input_type)
+        summed_out.div_(world_size)
+
+        summed_out_fp8, scale = cast_to_fp8(summed_out, fp8_format=fp8_format)
+
+        tensor_list_single = torch.empty(summed_out_fp8.size(0) * world_size, device=input_device, dtype=torch.uint8)
+        fut2 = dist.all_gather_into_tensor(
+            tensor_list_single, summed_out_fp8.view(torch.uint8), group=group_to_use, async_op=True
+        ).get_future()
+
+        scale_list = [torch.ones(1, dtype=scale.dtype, device=input_device) for _ in range(world_size)]
+        fut3 = dist.all_gather_into_tensor(
+            torch.cat(scale_list, dim=0), scale, group=group_to_use, async_op=True
+        ).get_future()
+        fut_combined2 = torch.futures.collect_all([fut2, fut3])
+        return fut_combined2
+
+    def decompress(fut):
+        tensor_list_single = fut.value().wait()[0].value()[0]
+        scale_list_single = fut.value().wait()[1].value()[0]
+
+        tensor_list = list(torch.chunk(tensor_list_single, world_size, dim=0))
+        scale_list = scale_list_single.chunk(world_size, dim=0)
+
+        for i in range(world_size):
+            tensor_list[i] = tensor_list[i].view(fp8_type).to(input_type) * scale_list[i]
+        out = torch.cat(tensor_list, dim=0)
+
+        input_tensor_size = input_tensor.numel()
+        input_shape = input_tensor.shape
+        out = out[:input_tensor_size]
+
+        input_tensor.copy_(out.view(input_shape).to(input_type))
+        return input_tensor
+
+    return all_to_all_fut.then(sum_and_allgather).then(decompress)
+
+
+def fp8_compress_ddp_grad_comm_hook_sync(
+    process_group: dist.ProcessGroup,
+    bucket: dist.GradBucket,
+    fp8_format="e5m2",
+) -> torch.futures.Future[torch.Tensor]:
+    """
+    Return a future that wraps the input, after the input is allreduced. However, the allreduce commnunication is synchronized.
+    This breaks the overlapping between allreduce communication and backward compuation.
+
+    This hook should **only** be used for debugging purposes, instead of the normal gradient synchronization.
+    For asynchronized implementation, use fp8_compress_ddp_grad_comm_hook_async instead.
+
+    Example::
+        >>> # xdoctest: +SKIP
+        >>> ddp_model.register_comm_hook(None, fp8_compress_ddp_grad_comm_hook_sync)
+    """
+
+    buffer = bucket.buffer()
+    all_reduce_fp8(buffer, fp8_format=fp8_format)
+
+    fut: torch.futures.Future[torch.Tensor] = torch.futures.Future()
+    fut.set_result(bucket.buffer())
+
+    return fut
+
+
+def fp8_compress_fsdp_grad_comm_hook(
+    state: object,
+    unsharded_gradient_flattened: torch.Tensor,
+    sharded_gradient: torch.Tensor,
+    group=None,
+    fp8_format="e5m2",
+) -> None:
+    """
+    This communication hook implements a simple gradient compression approach that casts unsharded_gradient_flattened tensor
+    to FP8 floating-point format (``torch.float8_e5m2`` or ``torch.bfloat16_e4m3``), and then perform scatter_allreduce logic
+    by using all_to_all and all_gather among the process group.
+
+    Example::
+        >>> fsdp_model.register_comm_hook(None, fp8_compress_fsdp_grad_comm_hook)
+    """
+    grad = unsharded_gradient_flattened
+    fp8_type = torch.float8_e4m3fn if fp8_format == "e4m3" else torch.float8_e5m2
+    input_type = grad.dtype
+    input_device = grad.device
+    world_size = dist.get_world_size(group=group)
+
+    grad_fp8, scale = cast_to_fp8(grad, fp8_format=fp8_format)
+    uint8_buffer = torch.empty_like(grad_fp8).view(torch.uint8)
+    dist.all_to_all_single(uint8_buffer, grad_fp8.view(torch.uint8), group=group)
+
+    scale_list = [torch.ones(1, dtype=scale.dtype, device=input_device) for _ in range(world_size)]
+    dist.all_gather(scale_list, scale, group=group)
+
+    buffer_list = list(torch.chunk(uint8_buffer.view(fp8_type), world_size, dim=0))
+    sharded_gradient.zero_()
+    for tensor, scale in zip(buffer_list, scale_list):
+        sharded_gradient += cast_from_fp8(tensor, scale, input_type)
+
+
+def fp8_compress_fsdp_params_comm_hook(
+    state: object,
+    padded_unsharded_flat_param: torch.Tensor,
+    sharded_flat_param: torch.Tensor,
+    group=None,
+    fp8_format="e5m2",
+) -> None:
+    """
+        This hook is pending the official support for parameters communication hook in FSDP, e.g. register_params_comm_hook.
+
+    Example::
+        >>> fsdp_model.register_params_comm_hook(None, fp8_compress_fsdp_params_comm_hook)
+    """
+
+    fp8_type = torch.float8_e4m3fn if fp8_format == "e4m3" else torch.float8_e5m2
+    fp8_max = torch.finfo(fp8_type).max
+    inp = sharded_flat_param
+    out = padded_unsharded_flat_param
+
+    per_tensor_max = inp.abs().max().float()
+    per_tensor_max = torch.where(per_tensor_max > 0, per_tensor_max, 1.0)
+    dist.all_reduce(per_tensor_max, op=torch.distributed.ReduceOp.MAX, group=group)
+
+    scale = fp8_max / per_tensor_max
+    fp8_sharded_flat_param = (scale * inp.float()).to(fp8_type).view(torch.uint8)
+
+    fp8_out = torch.empty(out.shape, dtype=torch.uint8, device=out.device)
+    dist.all_gather_into_tensor(
+        fp8_out,
+        fp8_sharded_flat_param,
+        group=group,
+    )
+    padded_unsharded_flat_param.copy_((fp8_out.view(fp8_type).float() / scale).to(out.dtype))
+
+
 def split_chunk_by_channel(
     chunk: torch.Tensor, channel_size: int, num_channels: int, rank: int = 0, world_size: int = 1
 ):
@@ -342,7 +543,7 @@ def all_gather_into_tensor_flat_fp8(
     scale_inv = 1.0 / scale
     buffer = torch.empty_like(output_tensor, dtype=fp8_type)
     dist.all_gather_into_tensor(buffer.view(torch.uint8), fp8_input.view(torch.uint8), group=group)
-    numel = np.prod(output_shape)
+    numel = output_shape.numel()
     valid_buffer = buffer[:numel].reshape(output_shape)
     valid_buffer = cast_from_fp8(valid_buffer, scale_inv, input_type, per_channel_scale=(len(output_shape) == 2))
     output_tensor[:numel].copy_(valid_buffer.view(-1))

colossalai/quantization/utils.py

@@ -0,0 +1,112 @@
+import torch
+import torch.distributed as dist
+from packaging import version
+from torch import Tensor
+from torch.distributed.fsdp._common_utils import _no_dispatch_record_stream
+from torch.distributed.utils import _p_assert
+
+
+def _all_gather_flat_param(
+    self,
+    padded_unsharded_flat_param: Tensor,
+) -> Tensor:
+    """
+    All-gather the handle's flat parameter to the destination ``padded_unsharded_flat_param``.
+
+    Then switch to use the all-gathered tensor.
+    """
+    _p_assert(
+        hasattr(self, "process_group") and hasattr(self, "world_size"),
+        "Expects a process group and world size to have been set via `shard()`",
+    )
+    sharded_flat_param = self.flat_param.data
+    expected_numel = sharded_flat_param.numel() * self.world_size
+    _p_assert(
+        padded_unsharded_flat_param.numel() == expected_numel,
+        f"Expects {expected_numel} numel but got {padded_unsharded_flat_param.numel()}",
+    )
+
+    pg = self._fake_process_group if self._use_fake_all_gather else self.process_group
+
+    # HACK this should be handled by C10D
+    if sharded_flat_param.is_cpu:  # type: ignore[attr-defined]
+        tensor_list = list(torch.chunk(padded_unsharded_flat_param, dist.get_world_size(pg)))
+        work = dist.all_gather(tensor_list, sharded_flat_param, group=pg)
+    else:
+        if self._comm_hook is None:
+            dist.all_gather_into_tensor(
+                padded_unsharded_flat_param,
+                sharded_flat_param,
+                pg,
+            )
+        else:
+            self._comm_hook(None, padded_unsharded_flat_param, sharded_flat_param, pg)
+
+    if self._offload_params:
+        # In case of offloading, `flat_param.data` (i.e. sharded param) is
+        # created on the pre-unshard stream. We need to hand it over to the
+        # unshard stream for all-gather
+        _no_dispatch_record_stream(
+            sharded_flat_param,
+            self._device_handle.current_stream(),  # unshard_stream
+        )
+    return padded_unsharded_flat_param
+
+
+def register_params_comm_hook(self, state: object, hook: callable):
+    """Register a communication hook for FlatParamHandle.
+
+    This is an enhancement that provides a flexible hook to users where they can specify how FSDP unshards
+    parameters across multiple workers.
+
+    .. warning ::
+        FSDP communication hook should be registered before running an initial forward pass
+        and only once.
+
+    Args:
+        state (object): Passed to the hook to maintain any state information during the training process.
+        hook (Callable): Callable, which has one of the following signatures:
+                        1) ``hook: Callable[torch.Tensor] -> None``:
+                        This function takes in a Python tensor, which represents
+                        the full, flattened, unsharded gradient with respect to all variables
+                        corresponding to the model this FSDP unit is wrapping
+                        (that are not wrapped by other FSDP sub-units).
+                        It then performs all necessary processing and returns ``None``;
+                        2) ``hook: Callable[torch.Tensor, torch.Tensor] -> None``:
+                        This function takes in two Python tensors, the first one represents
+                        the full, flattened, unsharded gradient with respect to all variables
+                        corresponding to the model this FSDP unit is wrapping
+                        (that are not wrapped by other FSDP sub-units). The latter
+                        represents a pre-sized tensor to store a chunk of a sharded gradient after
+                        reduction.
+                        In both cases, callable performs all necessary processing and returns ``None``.
+                        Callables with signature 1 are expected to handle gradient communication for a `NO_SHARD` case.
+                        Callables with signature 2 are expected to handle gradient communication for sharded cases.
+
+    """
+    if not self.check_is_root():
+        raise AssertionError("register_comm_hook can only be called on a root instance.")
+
+    # if fsdp_state.sharding_strategy in HYBRID_SHARDING_STRATEGIES:
+    #     raise AssertionError(
+    #         f"Communication hook is not supported for hybrid strategies: {fsdp_state.sharding_strategy}"
+    #     )
+    if self._handle._comm_hook is not None:
+        raise AssertionError("A communication hook is already registered")
+    if not callable(hook):
+        raise ValueError(f"The communication hook must be callable but got {hook}")
+    self._handle._comm_hook = hook
+    self._handle._comm_hook_state = state
+
+
+def patch_fsdp_params_comm_hook():
+    if version.parse(torch.__version__) >= version.parse("2.2.0"):
+        from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
+        from torch.distributed.fsdp._flat_param import FlatParamHandle
+
+        FlatParamHandle._comm_hook = None
+        FlatParamHandle._comm_hook_state = None
+        FlatParamHandle._all_gather_flat_param = _all_gather_flat_param
+        FSDP.register_params_comm_hook = register_params_comm_hook
+    else:
+        raise RuntimeError("This fsdp_params_comm_hook patch is not supported while torch version under 2.2.0.")

colossalai/zero/gemini/chunk/chunk.py

@@ -166,6 +166,7 @@ class Chunk:
         self.grad_chunk = None
         # the async all-reduce/reduce-scatter work of this grad chunk (None means sync)
         self.grad_reduce_work = None
+        self.fp8_communication = False
 
     @property
     def memory_usage(self) -> Dict[str, int]:
@@ -521,6 +522,14 @@ class Chunk:
 
             alloc_storage(self.cuda_global_chunk)
             assert self.cuda_global_chunk.is_contiguous()
+            if self.fp8_communication:
+                assert async_op == False, "fp8 all-gather does not support async_op!"
+                from colossalai.quantization.fp8 import all_gather_into_tensor_flat_fp8
+
+                work = all_gather_into_tensor_flat_fp8(
+                    self.cuda_global_chunk, self.cuda_shard, self.cuda_global_chunk.shape, self.torch_pg
+                )
+            else:
                 work = dist.all_gather_into_tensor(
                     self.cuda_global_chunk, self.cuda_shard, self.torch_pg, async_op=async_op
                 )

colossalai/zero/gemini/chunk/manager.py

@@ -26,6 +26,7 @@ class ChunkManager:
         init_device: Optional[torch.device] = None,
         reuse_fp16_chunk: bool = True,
         max_prefetch: int = 0,
+        fp8_communication: bool = False,
     ) -> None:
         self.device = init_device or get_accelerator().get_current_device()
         self.dp_degree_chunk_size_dict: Dict[int, int] = dict()
@@ -44,6 +45,7 @@ class ChunkManager:
         self.accumulating_grads = False
         self.overflow_counter = torch.tensor([0], dtype=torch.int, device=get_accelerator().get_current_device())
         self._prefetch_stream = get_accelerator().Stream() if max_prefetch else None
+        self.fp8_communication = fp8_communication
 
     def register_tensor(
         self,
@@ -101,6 +103,8 @@ class ChunkManager:
                 extra_dp_group=extra_dp_group,
                 **chunk_kwargs,
             )
+            if self.fp8_communication:
+                chunk.fp8_communication = True
 
             chunk_group.append(chunk)
             chunk.append_tensor(tensor)

colossalai/zero/gemini/gemini_ddp.py

@@ -98,6 +98,7 @@ class GeminiDDP(ModelWrapper):
         extra_dp_group: Optional[ProcessGroup] = None,
         verbose: bool = False,
         enable_async_reduce: bool = True,
+        fp8_communication: bool = False,
     ) -> None:
         assert mixed_precision in (torch.float16, torch.bfloat16)
         reuse_fp16_chunk = master_weights if not enable_gradient_accumulation else False
@@ -122,6 +123,8 @@ class GeminiDDP(ModelWrapper):
                 verbose=verbose,
                 max_prefetch=max_prefetch,
             )
+        if fp8_communication:
+            self.chunk_manager.fp8_communication = True
         self.gemini_manager = GeminiManager(
             placement_policy,
             self.chunk_manager,

examples/language/bert/finetune.py

@@ -179,7 +179,7 @@ def main():
         "--plugin",
         type=str,
         default="torch_ddp",
-        choices=["torch_ddp", "torch_ddp_fp16", "gemini", "low_level_zero", "hybrid_parallel"],
+        choices=["torch_ddp", "torch_ddp_fp16", "gemini", "low_level_zero", "hybrid_parallel", "torch_fsdp"],
         help="plugin to use",
     )
     parser.add_argument(
@@ -215,9 +215,9 @@ def main():
     if args.plugin == "torch_ddp_fp16":
         booster_kwargs["mixed_precision"] = "fp16"
     if args.plugin.startswith("torch_ddp"):
-        plugin = TorchDDPPlugin()
+        plugin = TorchDDPPlugin(fp8_communication=args.use_fp8_comm)
     elif args.plugin == "gemini":
-        plugin = GeminiPlugin(initial_scale=2**5)
+        plugin = GeminiPlugin(initial_scale=2**5, fp8_communication=args.use_fp8_comm)
     elif args.plugin == "low_level_zero":
         plugin = LowLevelZeroPlugin(initial_scale=2**5)
     elif args.plugin == "hybrid_parallel":
@@ -235,6 +235,17 @@ def main():
             initial_scale=1,
             fp8_communication=args.use_fp8_comm,
         )
+    elif args.plugin == "torch_fsdp":
+        from torch.distributed.fsdp.fully_sharded_data_parallel import MixedPrecision
+
+        from colossalai.booster.plugin import TorchFSDPPlugin
+
+        plugin = TorchFSDPPlugin(
+            mixed_precision=MixedPrecision(
+                param_dtype=torch.float16, reduce_dtype=torch.float16, buffer_dtype=torch.float16
+            ),
+            fp8_communication=args.use_fp8_comm,
+        )
 
     booster = Booster(plugin=plugin, **booster_kwargs)
 

examples/language/gpt/hybridparallelism/finetune.py

@@ -212,7 +212,7 @@ def main():
     if args.plugin == "torch_ddp_fp16":
         booster_kwargs["mixed_precision"] = "fp16"
     if args.plugin.startswith("torch_ddp"):
-        plugin = TorchDDPPlugin()
+        plugin = TorchDDPPlugin(fp8_communication=args.use_fp8_comm)
     elif args.plugin == "gemini":
         plugin = GeminiPlugin(initial_scale=2**5)
     elif args.plugin == "low_level_zero":

examples/language/llama/benchmark.py

@@ -98,7 +98,7 @@ def main():
     parser.add_argument("--disable-async-reduce", action="store_true", help="Disable the asynchronous reduce operation")
     parser.add_argument("--prefetch_num", type=int, default=0, help="chunk prefetch max number")
     parser.add_argument("--no_cache", action="store_true")
-    parser.add_argument("--overlap_allgather", action="store_true")
+    parser.add_argument("--use_fp8_comm", action="store_true", default=False, help="for using fp8 during communication")
     args = parser.parse_args()
 
     colossalai.launch_from_torch()
@@ -158,6 +158,7 @@ def main():
                     buffer_dtype=torch.float16,
                 ),
                 param_init_fn=empty_init(),
+                fp8_communication=args.use_fp8_comm,
             )
         else:
             plugin = TorchFSDPPlugin(
@@ -165,7 +166,8 @@ def main():
                     param_dtype=torch.float16,
                     reduce_dtype=torch.float16,
                     buffer_dtype=torch.float16,
-                )
+                ),
+                fp8_communication=args.use_fp8_comm,
             )
     elif args.plugin == "fsdp_cpu":
         if use_empty_init:
@@ -177,6 +179,7 @@ def main():
                 ),
                 cpu_offload=CPUOffload(offload_params=True),
                 param_init_fn=empty_init(),
+                fp8_communication=args.use_fp8_comm,
             )
         else:
             plugin = TorchFSDPPlugin(
@@ -186,6 +189,7 @@ def main():
                     buffer_dtype=torch.float16,
                 ),
                 cpu_offload=CPUOffload(offload_params=True),
+                fp8_communication=args.use_fp8_comm,
             )
     elif args.plugin == "3d":
         plugin = HybridParallelPlugin(
@@ -200,9 +204,9 @@ def main():
             enable_flash_attention=args.xformers,
             microbatch_size=args.mbs,
             precision="bf16",
+            dp_outside=False,
             overlap_p2p=args.overlap,
             enable_metadata_cache=not args.no_cache,
-            overlap_allgather=args.overlap_allgather,
             **hybrid_kwargs,
         )
     elif args.plugin == "3d_cpu":
@@ -293,7 +297,7 @@ def main():
     with get_profile_context(
         args.profile,
         args.ignore_steps,
-        1,  # avoid creating massive log files
+        len(dataloader) - 1,
         save_dir=f"profile/{time.strftime('%H:%M', time.localtime())}-{args.plugin}-llama-{args.config}",
     ) as prof:
         if isinstance(plugin, HybridParallelPlugin) and args.pp > 1:

tests/test_fp8/test_fp8_cast.py

@@ -0,0 +1,26 @@
+import torch
+from torch.testing import assert_close
+
+from colossalai.accelerator import get_accelerator
+from colossalai.quantization.fp8 import cast_from_fp8, cast_from_fp8_pipeline, cast_to_fp8, cast_to_fp8_pipeline
+from colossalai.testing import parameterize
+
+
+@parameterize("shape", [(100, 10), (10, 100), (3, 7), (2, 1), (1, 2), (2, 2), (4, 2), (5,), (4,), (2,)])
+@parameterize("dtype", [torch.bfloat16, torch.float16, torch.float32])
+@parameterize("fp8_format", ["e4m3", "e5m2"])
+def test_fp8_cast(shape, dtype, fp8_format):
+    x = torch.rand(shape, dtype=dtype, device=get_accelerator().get_current_device())
+    ret, scale_inv = cast_to_fp8(x, fp8_format=fp8_format)
+    out = cast_from_fp8(ret, scale_inv, x.dtype)
+    assert_close(out, x, rtol=0.1, atol=0.1)
+
+    if x.size(-1) % 2 == 0:
+        inp_dict = {"hidden_states": x.clone()}
+        cast_to_fp8_pipeline(inp_dict)
+        cast_from_fp8_pipeline(inp_dict)
+        assert_close(inp_dict["hidden_states"], x, rtol=0.1, atol=0.1)
+
+
+if __name__ == "__main__":
+    test_fp8_cast()

tests/test_fp8/test_fp8_ddp_comm_hook.py

@@ -0,0 +1,87 @@
+import os
+
+import torch
+import torch.distributed as dist
+import torch.multiprocessing as mp
+import torch.nn as nn
+import torch.optim as optim
+from torch.nn.parallel import DistributedDataParallel as DDP
+from torch.testing import assert_close
+
+# example modified from https://pytorch.org/tutorials/intermediate/ddp_tutorial.html
+
+
+def setup(rank, world_size):
+    os.environ["MASTER_ADDR"] = "localhost"
+    os.environ["MASTER_PORT"] = "12355"
+
+    # initialize the process group
+    dist.init_process_group("nccl", rank=rank, world_size=world_size)
+
+
+def cleanup():
+    dist.destroy_process_group()
+
+
+class ToyModel(nn.Module):
+    def __init__(self):
+        super(ToyModel, self).__init__()
+        self.net1 = nn.Linear(10, 10)
+        self.relu = nn.ReLU()
+        self.net2 = nn.Linear(10, 5)
+
+    def forward(self, x):
+        return self.net2(self.relu(self.net1(x)))
+
+
+def demo_basic(rank, world_size):
+    print(f"Running basic DDP example on rank {rank}.")
+    setup(rank, world_size)
+
+    def get_grads_after_one_iteration(hook=None):
+        torch.manual_seed(0)
+        # create model and move it to GPU with id rank
+        model = ToyModel().to(rank)
+
+        ddp_model = DDP(model, device_ids=[rank])
+
+        if hook is not None:
+            ddp_model.register_comm_hook(None, hook)
+
+        loss_fn = nn.MSELoss()
+        optimizer = optim.SGD(ddp_model.parameters(), lr=0.001)
+
+        optimizer.zero_grad()
+        outputs = ddp_model(torch.randn(20, 10))
+        labels = torch.randn(20, 5).to(rank)
+        loss_fn(outputs, labels).backward()
+        optimizer.step()
+
+        torch.distributed.barrier()
+
+        grad_dict = {}
+        for name, params in ddp_model.named_parameters():
+            grad_dict[name] = params.grad
+        return grad_dict
+
+    from colossalai.quantization.fp8 import fp8_compress_ddp_grad_comm_hook_async, fp8_compress_ddp_grad_comm_hook_sync
+
+    grad_dict = get_grads_after_one_iteration()
+    for hook in [fp8_compress_ddp_grad_comm_hook_sync, fp8_compress_ddp_grad_comm_hook_async]:
+        grad_dict_w_hook = get_grads_after_one_iteration(hook)
+        if dist.get_rank() == 0:
+            for name in grad_dict:
+                assert_close(grad_dict[name], grad_dict_w_hook[name], rtol=0.1, atol=0.1)
+
+    cleanup()
+
+
+def run_demo(demo_fn, world_size):
+    mp.spawn(demo_fn, args=(world_size,), nprocs=world_size, join=True)
+
+
+if __name__ == "__main__":
+    n_gpus = torch.cuda.device_count()
+    assert n_gpus >= 2, f"Requires at least 2 GPUs to run, but got {n_gpus}"
+    world_size = n_gpus
+    run_demo(demo_basic, world_size)

tests/test_fp8/test_fp8_fsdp_comm_hook.py

@@ -0,0 +1,107 @@
+import pytest
+import torch
+import torch.distributed as dist
+import torch.nn as nn
+import torch.optim as optim
+from packaging import version
+from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
+from torch.testing import assert_close
+
+from colossalai import launch
+from colossalai.testing import parameterize, rerun_if_address_is_in_use, spawn
+
+# example modified from https://pytorch.org/tutorials/intermediate/ddp_tutorial.html
+
+
+def cleanup():
+    dist.destroy_process_group()
+
+
+class ToyModel(nn.Module):
+    def __init__(self):
+        super(ToyModel, self).__init__()
+        self.net1 = nn.Linear(100, 100)
+        self.relu = nn.ReLU()
+        self.net2 = nn.Linear(100, 50)
+
+    def forward(self, x):
+        return self.net2(self.relu(self.net1(x)))
+
+
+@parameterize("mode", ["grad", "params"])
+def run_model(mode):
+    rank = dist.get_rank()
+
+    from colossalai.quantization.utils import patch_fsdp_params_comm_hook
+
+    patch_fsdp_params_comm_hook()
+
+    def get_grads_after_one_iteration(grad_hook=None, params_hook=None):
+        torch.manual_seed(0)
+        # create model and move it to GPU with id rank
+        model = ToyModel().to(rank)
+        fsdp_model = FSDP(model)
+
+        if grad_hook is not None:
+            fsdp_model.register_comm_hook(None, grad_hook)
+
+        if params_hook is not None:
+            fsdp_model.register_params_comm_hook(None, params_hook)
+
+        loss_fn = nn.MSELoss()
+        optimizer = optim.SGD(fsdp_model.parameters(), lr=0.001)
+
+        optimizer.zero_grad()
+        outputs = fsdp_model(torch.randn(20, 100))
+        labels = torch.randn(20, 50).to(rank)
+        loss_fn(outputs, labels).backward()
+        optimizer.step()
+
+        torch.distributed.barrier()
+
+        grad_dict = {}
+        for name, params in fsdp_model.named_parameters():
+            grad_dict[name] = params.grad
+        return grad_dict
+
+    from colossalai.quantization.fp8 import fp8_compress_fsdp_grad_comm_hook, fp8_compress_fsdp_params_comm_hook
+
+    if mode == "grad":
+        grad_dict = get_grads_after_one_iteration()
+        for hook in [
+            fp8_compress_fsdp_grad_comm_hook,
+        ]:
+            grad_dict_w_hook = get_grads_after_one_iteration(grad_hook=hook)
+            if dist.get_rank() == 0:
+                for name in grad_dict:
+                    assert_close(grad_dict[name], grad_dict_w_hook[name], rtol=0.1, atol=0.1)
+    elif mode == "params":
+        grad_dict = get_grads_after_one_iteration()
+        for hook in [
+            fp8_compress_fsdp_params_comm_hook,
+        ]:
+            grad_dict_w_hook = get_grads_after_one_iteration(params_hook=hook)
+            if dist.get_rank() == 0:
+                for name in grad_dict:
+                    assert_close(grad_dict[name], grad_dict_w_hook[name], rtol=0.1, atol=0.1)
+    else:
+        raise NotImplementedError
+
+
+def demo_basic(rank, world_size, port):
+    print(f"Running basic FSDP example on rank {rank}.")
+    launch(rank=rank, world_size=world_size, port=port, host="localhost")
+    run_model()
+    cleanup()
+
+
+@pytest.mark.skipif(version.parse(torch.__version__) < version.parse("2.2.0"), reason="torch version < 2.2.0.")
+@rerun_if_address_is_in_use()
+def test_fsdp():
+    n_gpus = torch.cuda.device_count()
+    assert n_gpus >= 2, f"Requires at least 2 GPUs to run, but got {n_gpus}"
+    spawn(demo_basic, n_gpus)
+
+
+if __name__ == "__main__":
+    test_fsdp()