fix sp i2v2

.github/workflows/pr-test.yml

@@ -122,17 +122,17 @@ jobs:
             # Actual tests
             encoder-test:
               - 'fastvideo/v1/models/encoders/**'
-              - 'fastvideo/v1/models/loader/**'
+              - 'fastvideo/v1/models/loaders/**'
               - 'fastvideo/v1/tests/encoders/**'
               - *common-paths
             vae-test:
               - 'fastvideo/v1/models/vaes/**'
-              - 'fastvideo/v1/models/loader/**'
+              - 'fastvideo/v1/models/loaders/**'
               - 'fastvideo/v1/tests/vaes/**'
               - *common-paths
             transformer-test:
               - 'fastvideo/v1/models/dits/**'
-              - 'fastvideo/v1/models/loader/**'
+              - 'fastvideo/v1/models/loaders/**'
               - 'fastvideo/v1/tests/transformers/**'
               - 'fastvideo/v1/layers/**'
               - 'fastvideo/v1/attention/**'
@@ -264,7 +264,7 @@ jobs:
     with:
       job_id: "training-test-VSA"
       gpu_type: "NVIDIA H100 NVL"
-      gpu_count: 1
+      gpu_count: 2
       volume_size: 100
       disk_size: 100
       image: "ghcr.io/${{ github.repository }}/fastvideo-dev:py3.12-latest"
@@ -284,7 +284,7 @@ jobs:
     with:
       job_id: "inference-test-STA"
       gpu_type: "NVIDIA H100 NVL"
-      gpu_count: 1
+      gpu_count: 2
       volume_size: 100
       disk_size: 100
       image: "ghcr.io/${{ github.repository }}/fastvideo-dev:py3.12-latest"

examples/inference/basic/basic.py

@@ -10,7 +10,7 @@ def main():
     # attempt to identify the optimal arguments.
     generator = VideoGenerator.from_pretrained(
         "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
-        # FastVideo will automatically handle distributed setup
+        # if num_gpus > 1, FastVideo will automatically handle distributed setup
         num_gpus=2,
         use_fsdp_inference=True,
         use_cpu_offload=False

fastvideo/v1/configs/models/base.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 from dataclasses import dataclass, field, fields
-from typing import Any, Dict, List, Tuple
+from typing import Any, Dict
 
 from fastvideo.v1.logger import init_logger
 
@@ -12,9 +12,7 @@ logger = init_logger(__name__)
 # 3. Any field in ArchConfig is fixed upon initialization, and should be hidden away from users
 @dataclass
 class ArchConfig:
-    stacked_params_mapping: List[Tuple[str, str, str]] = field(
-        default_factory=list
-    )  # mapping from huggingface weight names to custom names
+    pass
 
 
 @dataclass

fastvideo/v1/configs/models/dits/stepvideo.py

@@ -5,11 +5,13 @@ from typing import List, Optional, Tuple, Union
 from fastvideo.v1.configs.models.dits.base import DiTArchConfig, DiTConfig
 
 
+def is_blocks(n: str, m) -> bool:
+    return "blocks" in n and str.isdigit(n.split(".")[-1])
+
+
 @dataclass
 class StepVideoArchConfig(DiTArchConfig):
-    _fsdp_shard_conditions: list = field(
-        default_factory=lambda:
-        [lambda n, m: "transformer_blocks" in n and n.split(".")[-1].isdigit()])
+    _fsdp_shard_conditions: list = field(default_factory=lambda: [is_blocks])
 
     _param_names_mapping: dict = field(
         default_factory=lambda: {

fastvideo/v1/configs/models/encoders/base.py

@@ -32,11 +32,8 @@ class TextEncoderArchConfig(EncoderArchConfig):
     output_past: bool = True
     scalable_attention: bool = True
     tie_word_embeddings: bool = False
-    stacked_params_mapping: List[Tuple[str, str, str]] = field(
-        default_factory=list
-    )  # mapping from huggingface weight names to custom names
+
     tokenizer_kwargs: Dict[str, Any] = field(default_factory=dict)
-    _fsdp_shard_conditions: list = field(default_factory=lambda: [])
 
     def __post_init__(self) -> None:
         self.tokenizer_kwargs = {

fastvideo/v1/configs/models/encoders/clip.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 from dataclasses import dataclass, field
-from typing import List, Optional, Tuple
+from typing import Optional
 
 from fastvideo.v1.configs.models.encoders.base import (ImageEncoderArchConfig,
                                                        ImageEncoderConfig,
@@ -8,14 +8,6 @@ from fastvideo.v1.configs.models.encoders.base import (ImageEncoderArchConfig,
                                                        TextEncoderConfig)
 
 
-def _is_transformer_layer(n: str, m) -> bool:
-    return "layers" in n and str.isdigit(n.split(".")[-1])
-
-
-def _is_embeddings(n: str, m) -> bool:
-    return n.endswith("embeddings")
-
-
 @dataclass
 class CLIPTextArchConfig(TextEncoderArchConfig):
     vocab_size: int = 49408
@@ -35,15 +27,6 @@ class CLIPTextArchConfig(TextEncoderArchConfig):
     bos_token_id: int = 49406
     eos_token_id: int = 49407
     text_len: int = 77
-    stacked_params_mapping: List[Tuple[str, str,
-                                       str]] = field(default_factory=lambda: [
-                                           # (param_name, shard_name, shard_id)
-                                           ("qkv_proj", "q_proj", "q"),
-                                           ("qkv_proj", "k_proj", "k"),
-                                           ("qkv_proj", "v_proj", "v"),
-                                       ])
-    _fsdp_shard_conditions: list = field(
-        default_factory=lambda: [_is_transformer_layer, _is_embeddings])
 
 
 @dataclass

fastvideo/v1/configs/models/encoders/llama.py

@@ -1,23 +1,11 @@
 # SPDX-License-Identifier: Apache-2.0
 from dataclasses import dataclass, field
-from typing import List, Optional, Tuple
+from typing import Optional
 
 from fastvideo.v1.configs.models.encoders.base import (TextEncoderArchConfig,
                                                        TextEncoderConfig)
 
 
-def _is_transformer_layer(n: str, m) -> bool:
-    return "layers" in n and str.isdigit(n.split(".")[-1])
-
-
-def _is_embeddings(n: str, m) -> bool:
-    return n.endswith("embed_tokens")
-
-
-def _is_final_norm(n: str, m) -> bool:
-    return n.endswith("norm")
-
-
 @dataclass
 class LlamaArchConfig(TextEncoderArchConfig):
     vocab_size: int = 32000
@@ -44,18 +32,6 @@ class LlamaArchConfig(TextEncoderArchConfig):
     head_dim: Optional[int] = None
     hidden_state_skip_layer: int = 2
     text_len: int = 256
-    stacked_params_mapping: List[Tuple[str, str, str]] = field(
-        default_factory=lambda: [
-            # (param_name, shard_name, shard_id)
-            (".qkv_proj", ".q_proj", "q"),
-            (".qkv_proj", ".k_proj", "k"),
-            (".qkv_proj", ".v_proj", "v"),
-            (".gate_up_proj", ".gate_proj", 0),  # type: ignore
-            (".gate_up_proj", ".up_proj", 1),  # type: ignore
-        ])
-    _fsdp_shard_conditions: list = field(
-        default_factory=lambda:
-        [_is_transformer_layer, _is_embeddings, _is_final_norm])
 
 
 @dataclass

fastvideo/v1/configs/models/encoders/t5.py

@@ -1,23 +1,11 @@
 # SPDX-License-Identifier: Apache-2.0
 from dataclasses import dataclass, field
-from typing import List, Optional, Tuple
+from typing import Optional
 
 from fastvideo.v1.configs.models.encoders.base import (TextEncoderArchConfig,
                                                        TextEncoderConfig)
 
 
-def _is_transformer_layer(n: str, m) -> bool:
-    return "block" in n and str.isdigit(n.split(".")[-1])
-
-
-def _is_embeddings(n: str, m) -> bool:
-    return n.endswith("shared")
-
-
-def _is_final_layernorm(n: str, m) -> bool:
-    return n.endswith("final_layer_norm")
-
-
 @dataclass
 class T5ArchConfig(TextEncoderArchConfig):
     vocab_size: int = 32128
@@ -41,16 +29,6 @@ class T5ArchConfig(TextEncoderArchConfig):
     eos_token_id: int = 1
     classifier_dropout: float = 0.0
     text_len: int = 512
-    stacked_params_mapping: List[Tuple[str, str,
-                                       str]] = field(default_factory=lambda: [
-                                           # (param_name, shard_name, shard_id)
-                                           (".qkv_proj", ".q", "q"),
-                                           (".qkv_proj", ".k", "k"),
-                                           (".qkv_proj", ".v", "v"),
-                                       ])
-    _fsdp_shard_conditions: list = field(
-        default_factory=lambda:
-        [_is_transformer_layer, _is_embeddings, _is_final_layernorm])
 
     # Referenced from https://github.com/huggingface/transformers/blob/main/src/transformers/models/t5/configuration_t5.py
     def __post_init__(self):

fastvideo/v1/dataset/benchmarks/benchmark_parquet_dataset_iterable_style.py

@@ -11,7 +11,7 @@ from fastvideo.v1.dataset.parquet_dataset_iterable_style import (
     build_parquet_iterable_style_dataloader)
 from fastvideo.v1.distributed import get_world_rank
 from fastvideo.v1.distributed.parallel_state import (
-    cleanup_dist_env_and_memory, get_local_torch_device,
+    cleanup_dist_env_and_memory, get_torch_device,
     maybe_init_distributed_environment_and_model_parallel)
 from fastvideo.v1.logger import init_logger
 
@@ -148,8 +148,8 @@ def main() -> None:
                 break
 
             # Move data to device
-            latents = latents.to(get_local_torch_device())
-            embeddings = embeddings.to(get_local_torch_device())
+            latents = latents.to(get_torch_device())
+            embeddings = embeddings.to(get_torch_device())
 
             # Calculate actual batch size
             batch_size = latents.size(0)

fastvideo/v1/dataset/benchmarks/benchmark_parquet_dataset_map_style.py

@@ -13,7 +13,7 @@ from fastvideo.v1.dataset.parquet_dataset_map_style import (
     build_parquet_map_style_dataloader)
 from fastvideo.v1.distributed import get_world_rank
 from fastvideo.v1.distributed.parallel_state import (
-    cleanup_dist_env_and_memory, get_local_torch_device,
+    cleanup_dist_env_and_memory, get_torch_device,
     maybe_init_distributed_environment_and_model_parallel)
 from fastvideo.v1.logger import init_logger
 
@@ -165,8 +165,8 @@ def main() -> None:
                 break
 
             # Move data to device
-            latents = latents.to(get_local_torch_device())
-            embeddings = embeddings.to(get_local_torch_device())
+            latents = latents.to(get_torch_device())
+            embeddings = embeddings.to(get_torch_device())
 
             # Calculate actual batch size
             batch_size = latents.size(0)

fastvideo/v1/distributed/__init__.py

@@ -3,10 +3,10 @@
 from fastvideo.v1.distributed.communication_op import *
 from fastvideo.v1.distributed.parallel_state import (
     cleanup_dist_env_and_memory, get_dp_group, get_dp_rank, get_dp_world_size,
-    get_local_torch_device, get_sp_group, get_sp_parallel_rank,
-    get_sp_world_size, get_tp_group, get_tp_rank, get_tp_world_size,
-    get_world_group, get_world_rank, get_world_size,
-    init_distributed_environment, initialize_model_parallel,
+    get_sp_group, get_sp_parallel_rank, get_sp_world_size, get_torch_device,
+    get_tp_group, get_tp_rank, get_tp_world_size, get_world_group,
+    get_world_rank, get_world_size, init_distributed_environment,
+    initialize_model_parallel,
     maybe_init_distributed_environment_and_model_parallel,
     model_parallel_is_initialized)
 from fastvideo.v1.distributed.utils import *
@@ -40,5 +40,5 @@ __all__ = [
     "get_tp_world_size",
 
     # Get torch device
-    "get_local_torch_device",
+    "get_torch_device",
 ]

fastvideo/v1/distributed/parallel_state.py

@@ -36,7 +36,6 @@ from unittest.mock import patch
 
 import torch
 import torch.distributed
-import torch.distributed as dist
 from torch.distributed import Backend, ProcessGroup, ReduceOp
 
 import fastvideo.v1.envs as envs
@@ -693,7 +692,6 @@ class GroupCoordinator:
 
 
 _WORLD: Optional[GroupCoordinator] = None
-_NODE: Optional[GroupCoordinator] = None
 
 
 def get_world_group() -> GroupCoordinator:
@@ -701,11 +699,6 @@ def get_world_group() -> GroupCoordinator:
     return _WORLD
 
 
-def get_node_group() -> GroupCoordinator:
-    assert _NODE is not None, ("node group is not initialized")
-    return _NODE
-
-
 def init_world_group(ranks: List[int], local_rank: int,
                      backend: str) -> GroupCoordinator:
     return GroupCoordinator(
@@ -717,18 +710,6 @@ def init_world_group(ranks: List[int], local_rank: int,
     )
 
 
-def init_node_group(local_rank: int, backend: str):
-    cpu_group = get_world_group().cpu_group
-    node_ranks = same_node_ranks(cpu_group)
-    node_size = len(node_ranks)
-    all_node_ranks = [
-        list(range(i * node_size, (i + 1) * node_size))
-        for i in range(dist.get_world_size() // node_size)
-    ]
-    global _NODE
-    _NODE = init_model_parallel_group(all_node_ranks, local_rank, backend)
-
-
 def init_model_parallel_group(
     group_ranks: List[List[int]],
     local_rank: int,
@@ -801,8 +782,6 @@ def init_distributed_environment(
     else:
         assert _WORLD.world_size == torch.distributed.get_world_size(), (
             "world group already initialized with a different world size")
-    # Init a group for each node
-    init_node_group(local_rank, backend)
 
 
 _SP: Optional[GroupCoordinator] = None
@@ -925,7 +904,7 @@ def get_dp_rank() -> int:
     return get_dp_group().rank_in_group
 
 
-def get_local_torch_device() -> torch.device:
+def get_torch_device() -> torch.device:
     """Return the torch device for the current rank."""
     return torch.device(f"cuda:{envs.LOCAL_RANK}")
 
@@ -1042,22 +1021,17 @@ def cleanup_dist_env_and_memory(shutdown_ray: bool = False):
             "torch._C._host_emptyCache() only available in Pytorch >=2.5")
 
 
-def same_node_ranks(pg: Union[ProcessGroup, StatelessProcessGroup],
-                    source_rank: int = 0) -> List[int]:
+def in_the_same_node_as(pg: Union[ProcessGroup, StatelessProcessGroup],
+                        source_rank: int = 0) -> List[bool]:
     """
-    This is a collective operation that returns ranks that are in the same node
+    This is a collective operation that returns if each rank is in the same node
     as the source rank. It tests if processes are attached to the same
     memory system (shared access to shared memory).
-    Args:
-        pg: the global process group to test
-        source_rank: the rank to test against
-    Returns:
-        A list of ranks that are in the same node as the source rank.
     """
     if isinstance(pg, ProcessGroup):
         assert torch.distributed.get_backend(
             pg) != torch.distributed.Backend.NCCL, (
-                "same_node_ranks should be tested with a non-NCCL group.")
+                "in_the_same_node_as should be tested with a non-NCCL group.")
         # local rank inside the group
         rank = torch.distributed.get_rank(group=pg)
         world_size = torch.distributed.get_world_size(group=pg)
@@ -1129,7 +1103,7 @@ def same_node_ranks(pg: Union[ProcessGroup, StatelessProcessGroup],
             rank_data = pg.broadcast_obj(is_in_the_same_node, src=i)
             aggregated_data += rank_data
 
-    return [i for i, x in enumerate(aggregated_data.tolist()) if x == 1]
+    return [x == 1 for x in aggregated_data.tolist()]
 
 
 def initialize_tensor_parallel_group(

fastvideo/v1/fastvideo_args.py

@@ -58,10 +58,8 @@ class FastVideoArgs:
 
     output_type: str = "pil"
 
-    use_cpu_offload: bool = True  # For DiT
+    use_cpu_offload: bool = True
     use_fsdp_inference: bool = True
-    text_encoder_offload: bool = True
-    pin_cpu_memory: bool = True
 
     # STA (Sliding Tile Attention) parameters
     mask_strategy_file_path: Optional[str] = None
@@ -210,7 +208,7 @@ class FastVideoArgs:
             "--use-cpu-offload",
             action=StoreBoolean,
             help=
-            "Use CPU offload for DiT inference. Enable if run out of memory with FSDP.",
+            "Use CPU offload for model inference. Enable if run out of memory with FSDP.",
         )
         parser.add_argument(
             "--use-fsdp-inference",
@@ -218,19 +216,7 @@ class FastVideoArgs:
             help=
             "Use FSDP for inference by sharding the model weights. Latency is very low due to prefetch--enable if run out of memory.",
         )
-        parser.add_argument(
-            "--text-encoder-cpu-offload",
-            action=StoreBoolean,
-            help=
-            "Use CPU offload for text encoder. Enable if run out of memory.",
-        )
-        parser.add_argument(
-            "--pin-cpu-memory",
-            action=StoreBoolean,
-            help=
-            "Pin memory for CPU offload. Only added as a temp workaround if it throws \"CUDA error: invalid argument\". "
-            "Should be enabled in almost all cases",
-        )
+
         parser.add_argument(
             "--disable-autocast",
             action=StoreBoolean,

fastvideo/v1/layers/layernorm.py

@@ -6,7 +6,6 @@ from typing import Optional, Tuple, Union
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from torch.distributed.tensor import DTensor
 
 from fastvideo.v1.layers.custom_op import CustomOp
 
@@ -39,6 +38,12 @@ class RMSNorm(CustomOp):
         if self.has_weight:
             self.weight = nn.Parameter(self.weight)
 
+    # if we do fully_shard(model.layer_norm), and we call layer_form.forward_native(input) instead of layer_norm(input),
+    # we need to call model.layer_norm.register_fsdp_forward_method(model, "forward_native") to make sure fsdp2 hooks are triggered
+    # for mixed precision and cpu offloading
+
+    # the even better way might be fully_shard(model.layer_norm, mp_policy=, cpu_offloading=), and call model.layer_norm(input). everything should work out of the box
+    # because fsdp2 hooks will be triggered with model.layer_norm.__call__
     def forward_native(
         self,
         x: torch.Tensor,
@@ -71,12 +76,7 @@ class RMSNorm(CustomOp):
         x = x * torch.rsqrt(variance + self.variance_epsilon)
         x = x.to(orig_dtype)
         if self.has_weight:
-            # TODO(wenxuan): When using CPU offload, FSDP has a bug that doesn't unwrap DTensor in final_layer_norm.
-            # Report this
-            if isinstance(self.weight, DTensor):
-                x = x * self.weight.to(x.device).full_tensor()
-            else:
-                x = x * self.weight
+            x = x * self.weight
         if residual is None:
             return x
         else:

fastvideo/v1/models/dits/stepvideo.py

@@ -455,7 +455,10 @@ class StepVideoTransformerBlock(nn.Module):
 
 class StepVideoModel(BaseDiT):
     # (Optional) Keep the same attribute for compatibility with splitting, etc.
-    _fsdp_shard_conditions = StepVideoConfig()._fsdp_shard_conditions
+    _fsdp_shard_conditions = [
+        lambda n, m: "transformer_blocks" in n and n.split(".")[-1].isdigit(),
+        # lambda n, m: "pos_embed" in n  # If needed for the patch embedding.
+    ]
     _param_names_mapping = StepVideoConfig()._param_names_mapping
     _reverse_param_names_mapping = StepVideoConfig(
     )._reverse_param_names_mapping

fastvideo/v1/models/encoders/base.py

@@ -1,7 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 from abc import ABC, abstractmethod
-from dataclasses import field
-from typing import List, Optional, Tuple
+from typing import Optional, Tuple
 
 import torch
 from torch import nn
@@ -13,9 +12,6 @@ from fastvideo.v1.platforms import AttentionBackendEnum
 
 
 class TextEncoder(nn.Module, ABC):
-    _fsdp_shard_conditions: list = field(default_factory=lambda: [])
-    _stacked_params_mapping: List[Tuple[str, str,
-                                        str]] = field(default_factory=list)
     _supported_attention_backends: Tuple[
         AttentionBackendEnum,
         ...] = TextEncoderConfig()._supported_attention_backends
@@ -23,8 +19,6 @@ class TextEncoder(nn.Module, ABC):
     def __init__(self, config: TextEncoderConfig) -> None:
         super().__init__()
         self.config = config
-        self._fsdp_shard_conditions = config._fsdp_shard_conditions
-        self._stacked_params_mapping = config.arch_config.stacked_params_mapping
         if not self.supported_attention_backends:
             raise ValueError(
                 f"Subclass {self.__class__.__name__} must define _supported_attention_backends"

fastvideo/v1/models/encoders/clip.py

@@ -596,7 +596,12 @@ class CLIPVisionModel(ImageEncoder):
     #        ref: https://github.com/vllm-project/vllm/pull/7186#discussion_r1734163986
     def load_weights(self, weights: Iterable[Tuple[str,
                                                    torch.Tensor]]) -> Set[str]:
-
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            ("qkv_proj", "q_proj", "q"),
+            ("qkv_proj", "k_proj", "k"),
+            ("qkv_proj", "v_proj", "v"),
+        ]
         params_dict = dict(self.named_parameters())
         loaded_params: Set[str] = set()
         layer_count = len(self.vision_model.encoder.layers)
@@ -615,8 +620,7 @@ class CLIPVisionModel(ImageEncoder):
                 if layer_idx >= layer_count:
                     continue
 
-            for (param_name, weight_name,
-                 shard_id) in self.config.arch_config.stacked_params_mapping:
+            for (param_name, weight_name, shard_id) in stacked_params_mapping:
                 if weight_name not in name:
                     continue
                 name = name.replace(weight_name, param_name)

fastvideo/v1/models/encoders/llama.py

@@ -369,7 +369,14 @@ class LlamaModel(TextEncoder):
 
     def load_weights(self, weights: Iterable[Tuple[str,
                                                    torch.Tensor]]) -> Set[str]:
-
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            (".qkv_proj", ".q_proj", "q"),
+            (".qkv_proj", ".k_proj", "k"),
+            (".qkv_proj", ".v_proj", "v"),
+            (".gate_up_proj", ".gate_proj", 0),
+            (".gate_up_proj", ".up_proj", 1),
+        ]
         params_dict = dict(self.named_parameters())
         loaded_params: Set[str] = set()
         for name, loaded_weight in weights:
@@ -399,7 +406,7 @@ class LlamaModel(TextEncoder):
                     continue
                 else:
                     name = kv_scale_name
-            for param_name, weight_name, shard_id in self.config.arch_config.stacked_params_mapping:
+            for param_name, weight_name, shard_id in stacked_params_mapping:
                 if weight_name not in name:
                     continue
                 name = name.replace(weight_name, param_name)

fastvideo/v1/models/encoders/t5.py

@@ -494,7 +494,7 @@ class T5Stack(nn.Module):
                 attention_mask=attention_mask,
                 attn_metadata=attn_metadata,
             )
-        hidden_states = self.final_layer_norm.forward(hidden_states)
+        hidden_states = self.final_layer_norm.forward_native(hidden_states)
         return hidden_states
 
 
@@ -631,13 +631,19 @@ class UMT5EncoderModel(TextEncoder):
 
     def load_weights(self, weights: Iterable[Tuple[str,
                                                    torch.Tensor]]) -> Set[str]:
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            (".qkv_proj", ".q", "q"),
+            (".qkv_proj", ".k", "k"),
+            (".qkv_proj", ".v", "v"),
+        ]
         params_dict = dict(self.named_parameters())
         loaded_params: Set[str] = set()
         for name, loaded_weight in weights:
             loaded = False
             if "decoder" in name or "lm_head" in name:
                 continue
-            for param_name, weight_name, shard_id in self.config.arch_config.stacked_params_mapping:
+            for param_name, weight_name, shard_id in stacked_params_mapping:
                 if weight_name not in name:
                     continue
                 name = name.replace(weight_name, param_name)

fastvideo/v1/models/loader/component_loader.py

@@ -10,20 +10,17 @@ from copy import deepcopy
 from typing import Any, Generator, Iterable, List, Optional, Tuple, cast
 
 import torch
-import torch.distributed as dist
 import torch.nn as nn
 from safetensors.torch import load_file as safetensors_load_file
 from transformers import AutoImageProcessor, AutoTokenizer
 from transformers.utils import SAFE_WEIGHTS_INDEX_NAME
 
 from fastvideo.v1.configs.models import EncoderConfig
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.models.hf_transformer_utils import get_diffusers_config
-from fastvideo.v1.models.loader.fsdp_load import (init_device_mesh,
-                                                  maybe_load_fsdp_model,
-                                                  shard_model)
+from fastvideo.v1.models.loader.fsdp_load import maybe_load_fsdp_model
 from fastvideo.v1.models.loader.utils import set_default_torch_dtype
 from fastvideo.v1.models.loader.weight_utils import (
     filter_duplicate_safetensors_files, filter_files_not_needed_for_inference,
@@ -166,19 +163,16 @@ class TextEncoderLoader(ComponentLoader):
         return hf_folder, hf_weights_files, use_safetensors
 
     def _get_weights_iterator(
-            self,
-            source: "Source",
-            to_cpu: bool = True
+            self, source: "Source"
     ) -> Generator[Tuple[str, torch.Tensor], None, None]:
         """Get an iterator for the model weights based on the load format."""
         hf_folder, hf_weights_files, use_safetensors = self._prepare_weights(
             source.model_or_path, source.fall_back_to_pt,
             source.allow_patterns_overrides)
         if use_safetensors:
-            weights_iterator = safetensors_weights_iterator(
-                hf_weights_files, to_cpu)
+            weights_iterator = safetensors_weights_iterator(hf_weights_files)
         else:
-            weights_iterator = pt_weights_iterator(hf_weights_files, to_cpu)
+            weights_iterator = pt_weights_iterator(hf_weights_files)
 
         if self.counter_before_loading_weights == 0.0:
             self.counter_before_loading_weights = time.perf_counter()
@@ -187,11 +181,10 @@ class TextEncoderLoader(ComponentLoader):
                 for (name, tensor) in weights_iterator)
 
     def _get_all_weights(
-            self,
-            model_config: Any,
-            model: nn.Module,
-            model_path: str,
-            to_cpu: bool = True
+        self,
+        model_config: Any,
+        model: nn.Module,
+        model_path: str,
     ) -> Generator[Tuple[str, torch.Tensor], None, None]:
         primary_weights = TextEncoderLoader.Source(
             model_path,
@@ -200,14 +193,14 @@ class TextEncoderLoader(ComponentLoader):
             allow_patterns_overrides=getattr(model, "allow_patterns_overrides",
                                              None),
         )
-        yield from self._get_weights_iterator(primary_weights, to_cpu)
+        yield from self._get_weights_iterator(primary_weights)
 
         secondary_weights = cast(
             Iterable[TextEncoderLoader.Source],
             getattr(model, "secondary_weights", ()),
         )
         for source in secondary_weights:
-            yield from self._get_weights_iterator(source, to_cpu)
+            yield from self._get_weights_iterator(source)
 
     def load(self, model_path: str, architecture: str,
              fastvideo_args: FastVideoArgs):
@@ -240,22 +233,16 @@ class TextEncoderLoader(ComponentLoader):
             encoder_precision = fastvideo_args.pipeline_config.text_encoder_precisions[
                 1]
 
-        target_device = get_local_torch_device()
+        target_device = get_torch_device()
         # TODO(will): add support for other dtypes
         return self.load_model(model_path, encoder_config, target_device,
-                               fastvideo_args, encoder_precision)
+                               encoder_precision)
 
     def load_model(self,
                    model_path: str,
                    model_config: EncoderConfig,
                    target_device: torch.device,
-                   fastvideo_args: FastVideoArgs,
                    dtype: str = "fp16"):
-        use_cpu_offload = fastvideo_args.text_encoder_offload and len(
-            getattr(model_config, "_fsdp_shard_conditions", [])) > 0
-
-        if fastvideo_args.text_encoder_offload:
-            target_device = torch.device("cpu")
         with set_default_torch_dtype(PRECISION_TO_TYPE[dtype]):
             with target_device:
                 architectures = getattr(model_config, "architectures", [])
@@ -264,26 +251,12 @@ class TextEncoderLoader(ComponentLoader):
 
             weights_to_load = {name for name, _ in model.named_parameters()}
             loaded_weights = model.load_weights(
-                self._get_all_weights(model_config, model, model_path,
-                                      use_cpu_offload))
+                self._get_all_weights(model_config, model, model_path))
             self.counter_after_loading_weights = time.perf_counter()
             logger.info(
                 "Loading weights took %.2f seconds",
                 self.counter_after_loading_weights -
                 self.counter_before_loading_weights)
-
-            if use_cpu_offload:
-                mesh = init_device_mesh(
-                    "cuda",
-                    mesh_shape=(1, dist.get_world_size()),
-                    mesh_dim_names=("offload", "replicate"),
-                )
-                shard_model(model,
-                            cpu_offload=True,
-                            reshard_after_forward=True,
-                            mesh=mesh["offload"],
-                            fsdp_shard_conditions=model._fsdp_shard_conditions,
-                            pin_cpu_memory=fastvideo_args.pin_cpu_memory)
             # We only enable strict check for non-quantized models
             # that have loaded weights tracking currently.
             # if loaded_weights is not None:
@@ -317,10 +290,10 @@ class ImageEncoderLoader(TextEncoderLoader):
         encoder_config = fastvideo_args.pipeline_config.image_encoder_config
         encoder_config.update_model_arch(model_config)
 
-        target_device = get_local_torch_device()
+        target_device = get_torch_device()
         # TODO(will): add support for other dtypes
         return self.load_model(
-            model_path, encoder_config, target_device, fastvideo_args,
+            model_path, encoder_config, target_device,
             fastvideo_args.pipeline_config.image_encoder_precision)
 
 
@@ -373,7 +346,7 @@ class VAELoader(ComponentLoader):
         with set_default_torch_dtype(PRECISION_TO_TYPE[
                 fastvideo_args.pipeline_config.vae_precision]):
             vae_cls, _ = ModelRegistry.resolve_model_cls(class_name)
-            vae = vae_cls(vae_config).to(get_local_torch_device())
+            vae = vae_cls(vae_config).to(get_torch_device())
 
         # Find all safetensors files
         safetensors_list = glob.glob(
@@ -432,7 +405,7 @@ class TransformerLoader(ComponentLoader):
                 "hf_config": hf_config
             },
             weight_dir_list=safetensors_list,
-            device=get_local_torch_device(),
+            device=get_torch_device(),
             hsdp_replicate_dim=fastvideo_args.hsdp_replicate_dim,
             hsdp_shard_dim=fastvideo_args.hsdp_shard_dim,
             cpu_offload=fastvideo_args.use_cpu_offload,

fastvideo/v1/models/loader/fsdp_load.py

@@ -69,7 +69,6 @@ def maybe_load_fsdp_model(
     fsdp_inference: bool = False,
     output_dtype: Optional[torch.dtype] = None,
     training_mode: bool = True,
-    pin_cpu_memory: bool = True,
 ) -> torch.nn.Module:
     """
     Load the model with FSDP if is training, else load the model without FSDP.
@@ -102,12 +101,9 @@ def maybe_load_fsdp_model(
                 cpu_offload=cpu_offload,
                 reshard_after_forward=True,
                 mp_policy=mp_policy,
-                mesh=device_mesh,
-                fsdp_shard_conditions=model._fsdp_shard_conditions,
-                pin_cpu_memory=pin_cpu_memory)
+                mesh=device_mesh)
 
-    weight_iterator = safetensors_weights_iterator(
-        weight_dir_list, to_cpu=cpu_offload, async_broadcast=not cpu_offload)
+    weight_iterator = safetensors_weights_iterator(weight_dir_list)
     param_names_mapping_fn = get_param_names_mapping(model._param_names_mapping)
     load_model_from_full_model_state_dict(
         model,
@@ -130,13 +126,12 @@ def maybe_load_fsdp_model(
 
 def shard_model(
     model,
+    *,
     cpu_offload: bool,
     reshard_after_forward: bool = True,
-    mp_policy: Optional[MixedPrecisionPolicy] = MixedPrecisionPolicy(),  # noqa
+    mp_policy: Optional[MixedPrecisionPolicy] = None,
+    dp_mesh: Optional[DeviceMesh] = None,
     mesh: Optional[DeviceMesh] = None,
-    fsdp_shard_conditions: Optional[List[Callable[[str, nn.Module],
-                                                  bool]]] = None,
-    pin_cpu_memory: bool = True,
 ) -> None:
     """
     Utility to shard a model with FSDP using the PyTorch Distributed fully_shard API.
@@ -155,28 +150,19 @@ def shard_model(
         reshard_after_forward (bool): Whether to reshard parameters and buffers after
             the forward pass. Setting this to True corresponds to the FULL_SHARD sharding strategy
             from FSDP1, while setting it to False corresponds to the SHARD_GRAD_OP sharding strategy.
-        mesh (Optional[DeviceMesh]): Device mesh to use for FSDP sharding under multiple parallelism.
+        dp_mesh (Optional[DeviceMesh]): Device mesh to use for FSDP sharding under multiple parallelism.
             Default to None.
-        fsdp_shard_conditions (Optional[List[Callable[[str, nn.Module], bool]]]): A list of functions to determine
-            which modules to shard with FSDP. 
 
     Raises:
         ValueError: If no layer modules were sharded, indicating that no shard_condition was triggered.
     """
-    if fsdp_shard_conditions is None or len(fsdp_shard_conditions) == 0:
-        logger.warning(
-            "The FSDP shard condition list is empty or None. No modules will be sharded in %s",
-            type(model).__name__)
-        return
-
     fsdp_kwargs = {
         "reshard_after_forward": reshard_after_forward,
         "mesh": mesh,
         "mp_policy": mp_policy,
     }
     if cpu_offload:
-        fsdp_kwargs["offload_policy"] = CPUOffloadPolicy(
-            pin_memory=pin_cpu_memory)
+        fsdp_kwargs["offload_policy"] = CPUOffloadPolicy()
 
     # iterating in reverse to start with
     # lowest-level modules first
@@ -186,7 +172,7 @@ def shard_model(
     for n, m in reversed(list(model.named_modules())):
         if any([
                 shard_condition(n, m)
-                for shard_condition in fsdp_shard_conditions
+                for shard_condition in model._fsdp_shard_conditions
         ]):
             fully_shard(m, **fsdp_kwargs)
             num_layers_sharded += 1
@@ -195,6 +181,7 @@ def shard_model(
         raise ValueError(
             "No layer modules were sharded. Please check if shard conditions are working as expected."
         )
+
     # Finally shard the entire model to account for any stragglers
     fully_shard(model, **fsdp_kwargs)
 
@@ -237,9 +224,6 @@ def load_model_from_full_model_state_dict(
     to_merge_params: DefaultDict[str, Dict[Any, Any]] = defaultdict(dict)
     reverse_param_names_mapping = {}
     assert param_names_mapping is not None
-
-    # iterate over all the weights to sync broadcast before use
-    full_sd_iterator = list(full_sd_iterator)  # type: ignore
     for source_param_name, full_tensor in full_sd_iterator:
         target_param_name, merge_index, num_params_to_merge = param_names_mapping(
             source_param_name)

fastvideo/v1/models/loader/weight_utils.py

@@ -11,11 +11,9 @@ from typing import Generator, List, Optional, Tuple, Union
 import filelock
 import huggingface_hub.constants
 import torch
-import torch.distributed as dist
 from safetensors.torch import safe_open
 from tqdm.auto import tqdm
 
-from fastvideo.v1.distributed.parallel_state import get_node_group
 from fastvideo.v1.logger import init_logger
 
 logger = init_logger(__name__)
@@ -120,77 +118,36 @@ _BAR_FORMAT = "{desc}: {percentage:3.0f}% Completed | {n_fmt}/{total_fmt} [{elap
 
 
 def safetensors_weights_iterator(
-    hf_weights_files: List[str],
-    to_cpu: bool = False,
-    async_broadcast: bool = False
+    hf_weights_files: List[str]
 ) -> Generator[Tuple[str, torch.Tensor], None, None]:
-    """Iterate over the weights in the model safetensor files.
-    Args:
-        hf_weights_files: List of safetensor files to load.
-        to_cpu: Whether to load the weights to CPU. If False, will load to the GPU device bound to the current process.
-        async_broadcast: Whether to overlap loading from disk and broadcasting to other ranks. If True,
-            must iterate over all the weights before use. Only use if to_cpu is False.
-    """
-    local_rank = get_node_group().rank
-    device = f"cuda:{local_rank}" if not to_cpu else "cpu"
-    enable_tqdm = not torch.distributed.is_initialized() or get_node_group(
-    ).rank == 0
-    assert not (async_broadcast
-                and to_cpu), "Cannot broadcast weights when loading to CPU"
-
-    handles = []
+    """Iterate over the weights in the model safetensor files."""
+    enable_tqdm = not torch.distributed.is_initialized(
+    ) or torch.distributed.get_rank() == 0
     for st_file in tqdm(
             hf_weights_files,
             desc="Loading safetensors checkpoint shards",
             disable=not enable_tqdm,
             bar_format=_BAR_FORMAT,
     ):
-        with safe_open(st_file, framework="pt", device=device) as f:
+        with safe_open(st_file, framework="pt") as f:
             for name in f.keys():  # noqa: SIM118
-                if to_cpu:
-                    param = f.get_tensor(name)
-                else:
-                    if local_rank == 0:
-                        param = f.get_tensor(name)
-                    else:
-                        shape = f.get_slice(name).get_shape()
-                        param = torch.empty(shape, device=device)
-                    # broadcast to local ranks
-                    # TODO(Wenxuan): scatter instead of broadcast
-                    if get_node_group().world_size > 1:
-                        group = get_node_group().device_group
-                        if async_broadcast:
-                            handle = dist.broadcast(param,
-                                                    src=dist.get_global_rank(
-                                                        group, 0),
-                                                    async_op=True)
-                            handles.append(handle)
-                        else:
-                            dist.broadcast(param,
-                                           src=dist.get_global_rank(group, 0))
+                param = f.get_tensor(name)
                 yield name, param
 
-        if async_broadcast:
-            for handle in handles:
-                handle.wait()
-
 
 def pt_weights_iterator(
-    hf_weights_files: List[str],
-    to_cpu: bool = True  # default to CPU for text encoder
+    hf_weights_files: List[str]
 ) -> Generator[Tuple[str, torch.Tensor], None, None]:
     """Iterate over the weights in the model bin/pt files."""
-    local_rank = get_node_group().rank
-    device = f"cuda:{local_rank}" if not to_cpu else "cpu"
-    enable_tqdm = not torch.distributed.is_initialized() or get_node_group(
-    ).rank == 0
+    enable_tqdm = not torch.distributed.is_initialized(
+    ) or torch.distributed.get_rank() == 0
     for bin_file in tqdm(
             hf_weights_files,
             desc="Loading pt checkpoint shards",
             disable=not enable_tqdm,
             bar_format=_BAR_FORMAT,
     ):
-        state = torch.load(bin_file, map_location=device, weights_only=True)
+        state = torch.load(bin_file, map_location="cpu", weights_only=True)
         yield from state.items()
         del state
 

fastvideo/v1/pipelines/preprocess/preprocess_pipeline_base.py

@@ -18,7 +18,7 @@ from tqdm import tqdm
 from fastvideo.v1.configs.sample import SamplingParam
 from fastvideo.v1.dataset import ValidationDataset, getdataset
 from fastvideo.v1.dataset.preprocessing_datasets import PreprocessBatch
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.pipelines.composed_pipeline_base import ComposedPipelineBase
@@ -328,8 +328,7 @@ class BasePreprocessPipeline(ComposedPipelineBase):
                 # VAE
                 with torch.autocast("cuda", dtype=torch.float32):
                     latents = self.get_module("vae").encode(
-                        valid_data["pixel_values"].to(
-                            get_local_torch_device())).mean
+                        valid_data["pixel_values"].to(get_torch_device())).mean
 
                 # Get extra features if needed
                 extra_features = self.get_extra_features(

fastvideo/v1/pipelines/preprocess/preprocess_pipeline_i2v.py

@@ -13,7 +13,7 @@ import torch
 from PIL import Image
 
 from fastvideo.v1.dataset.dataloader.schema import pyarrow_schema_i2v
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.forward_context import set_forward_context
 from fastvideo.v1.models.vision_utils import (get_default_height_width,
@@ -82,8 +82,8 @@ class PreprocessPipeline_I2V(BasePreprocessPipeline):
                            fastvideo_args: FastVideoArgs) -> Dict[str, Any]:
 
         # TODO(will): move these to cpu at some point
-        self.get_module("image_encoder").to(get_local_torch_device())
-        self.get_module("vae").to(get_local_torch_device())
+        self.get_module("image_encoder").to(get_torch_device())
+        self.get_module("vae").to(get_torch_device())
 
         features = {}
         """Get CLIP features from the first frame of each video."""
@@ -107,7 +107,7 @@ class PreprocessPipeline_I2V(BasePreprocessPipeline):
         # Get CLIP features
         pixel_values = torch.cat(
             [img['pixel_values'] for img in processed_images],
-            dim=0).to(get_local_torch_device())
+            dim=0).to(get_torch_device())
         with torch.no_grad():
             image_inputs = {'pixel_values': pixel_values}
             with set_forward_context(current_timestep=0, attn_metadata=None):
@@ -129,8 +129,8 @@ class PreprocessPipeline_I2V(BasePreprocessPipeline):
                                         height, width)
             ],
                                         dim=2)
-            video_condition = video_condition.to(
-                device=get_local_torch_device(), dtype=torch.float32)
+            video_condition = video_condition.to(device=get_torch_device(),
+                                                 dtype=torch.float32)
             video_conditions.append(video_condition)
 
         video_conditions = torch.cat(video_conditions, dim=0)

fastvideo/v1/pipelines/stages/decoding.py

@@ -5,7 +5,7 @@ Decoding stage for diffusion pipelines.
 
 import torch
 
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.models.vaes.common import ParallelTiledVAE
@@ -61,7 +61,7 @@ class DecodingStage(PipelineStage):
         Returns:
             The batch with decoded outputs.
         """
-        self.vae = self.vae.to(get_local_torch_device())
+        self.vae = self.vae.to(get_torch_device())
 
         latents = batch.latents
         # TODO(will): remove this once we add input/output validation for stages

fastvideo/v1/pipelines/stages/denoising.py

@@ -12,9 +12,8 @@ from tqdm.auto import tqdm
 
 from fastvideo.v1.attention import get_attn_backend
 from fastvideo.v1.configs.pipelines.base import STA_Mode
-from fastvideo.v1.distributed import (get_local_torch_device,
-                                      get_sp_parallel_rank, get_sp_world_size,
-                                      get_world_group)
+from fastvideo.v1.distributed import (get_sp_parallel_rank, get_sp_world_size,
+                                      get_torch_device, get_world_group)
 from fastvideo.v1.distributed.communication_op import (
     sequence_model_parallel_all_gather)
 from fastvideo.v1.fastvideo_args import FastVideoArgs
@@ -193,7 +192,7 @@ class DenoisingStage(PipelineStage):
                         [fastvideo_args.pipeline_config.embedded_cfg_scale] *
                         latent_model_input.shape[0],
                         dtype=torch.float32,
-                        device=get_local_torch_device(),
+                        device=get_torch_device(),
                     ).to(target_dtype) *
                     1000.0 if fastvideo_args.pipeline_config.embedded_cfg_scale
                     is not None else None)

fastvideo/v1/pipelines/stages/encoding.py

@@ -7,7 +7,7 @@ from typing import Optional
 import PIL.Image
 import torch
 
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.models.vaes.common import ParallelTiledVAE
@@ -49,7 +49,7 @@ class EncodingStage(PipelineStage):
         Returns:
             The batch with encoded outputs.
         """
-        self.vae = self.vae.to(get_local_torch_device())
+        self.vae = self.vae.to(get_torch_device())
 
         assert batch.height is not None
         assert batch.width is not None
@@ -65,8 +65,7 @@ class EncodingStage(PipelineStage):
                 image,
                 vae_scale_factor=self.vae.spatial_compression_ratio,
                 height=batch.height,
-                width=batch.width).to(get_local_torch_device(),
-                                      dtype=torch.float32)
+                width=batch.width).to(get_torch_device(), dtype=torch.float32)
 
             image = image.unsqueeze(2)
         else:
@@ -79,7 +78,7 @@ class EncodingStage(PipelineStage):
                             batch.num_frames - 1, batch.height, batch.width)
         ],
                                     dim=2)
-        video_condition = video_condition.to(device=get_local_torch_device(),
+        video_condition = video_condition.to(device=get_torch_device(),
                                              dtype=torch.float32)
 
         # Setup VAE precision

fastvideo/v1/pipelines/stages/image_encoding.py

@@ -7,7 +7,7 @@ This module contains implementations of image encoding stages for diffusion pipe
 
 import torch
 
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.forward_context import set_forward_context
 from fastvideo.v1.logger import init_logger
@@ -55,12 +55,12 @@ class ImageEncodingStage(PipelineStage):
             The batch with encoded prompt embeddings.
         """
         if fastvideo_args.use_cpu_offload:
-            self.image_encoder = self.image_encoder.to(get_local_torch_device())
+            self.image_encoder = self.image_encoder.to(get_torch_device())
 
         image = batch.pil_image
 
         image_inputs = self.image_processor(
-            images=image, return_tensors="pt").to(get_local_torch_device())
+            images=image, return_tensors="pt").to(get_torch_device())
         with set_forward_context(current_timestep=0, attn_metadata=None):
             outputs = self.image_encoder(**image_inputs)
             image_embeds = outputs.last_hidden_state

fastvideo/v1/pipelines/stages/latent_preparation.py

@@ -5,7 +5,7 @@ Latent preparation stage for diffusion pipelines.
 
 from diffusers.utils.torch_utils import randn_tensor
 
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.pipelines.pipeline_batch_info import ForwardBatch
@@ -62,7 +62,7 @@ class LatentPreparationStage(PipelineStage):
 
         # Get required parameters
         dtype = batch.prompt_embeds[0].dtype
-        device = get_local_torch_device()
+        device = get_torch_device()
         generator = batch.generator
         latents = batch.latents
         num_frames = latent_num_frames if latent_num_frames is not None else batch.num_frames

fastvideo/v1/pipelines/stages/text_encoding.py

@@ -5,7 +5,9 @@ Prompt encoding stages for diffusion pipelines.
 This module contains implementations of prompt encoding stages for diffusion pipelines.
 """
 
-from fastvideo.v1.distributed import get_local_torch_device
+import torch
+
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.forward_context import set_forward_context
 from fastvideo.v1.pipelines.pipeline_batch_info import ForwardBatch
@@ -60,6 +62,8 @@ class TextEncodingStage(PipelineStage):
                 fastvideo_args.pipeline_config.text_encoder_configs,
                 fastvideo_args.pipeline_config.preprocess_text_funcs,
                 fastvideo_args.pipeline_config.postprocess_text_funcs):
+            if fastvideo_args.use_cpu_offload:
+                text_encoder = text_encoder.to(get_torch_device())
 
             assert isinstance(batch.prompt, (str, list))
             if isinstance(batch.prompt, str):
@@ -67,9 +71,8 @@ class TextEncodingStage(PipelineStage):
             texts = []
             for prompt_str in batch.prompt:
                 texts.append(preprocess_func(prompt_str))
-            text_inputs = tokenizer(texts,
-                                    **encoder_config.tokenizer_kwargs).to(
-                                        get_local_torch_device())
+            text_inputs = tokenizer(
+                texts, **encoder_config.tokenizer_kwargs).to(get_torch_device())
             input_ids = text_inputs["input_ids"]
             attention_mask = text_inputs["attention_mask"]
             with set_forward_context(current_timestep=0, attn_metadata=None):
@@ -88,8 +91,8 @@ class TextEncodingStage(PipelineStage):
                 assert isinstance(batch.negative_prompt, str)
                 negative_text = preprocess_func(batch.negative_prompt)
                 negative_text_inputs = tokenizer(
-                    negative_text, **encoder_config.tokenizer_kwargs).to(
-                        get_local_torch_device())
+                    negative_text,
+                    **encoder_config.tokenizer_kwargs).to(get_torch_device())
                 negative_input_ids = negative_text_inputs["input_ids"]
                 negative_attention_mask = negative_text_inputs["attention_mask"]
                 with set_forward_context(current_timestep=0,
@@ -107,8 +110,9 @@ class TextEncodingStage(PipelineStage):
                     batch.negative_attention_mask.append(
                         negative_attention_mask)
 
-            if fastvideo_args.text_encoder_offload:
+            if fastvideo_args.use_cpu_offload:
                 text_encoder.to('cpu')
+                torch.cuda.empty_cache()
 
         return batch
 

fastvideo/v1/pipelines/stages/timestep_preparation.py

@@ -7,7 +7,7 @@ This module contains implementations of timestep preparation stages for diffusio
 
 import inspect
 
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.pipelines.pipeline_batch_info import ForwardBatch
@@ -45,7 +45,7 @@ class TimestepPreparationStage(PipelineStage):
             The batch with prepared timesteps.
         """
         scheduler = self.scheduler
-        device = get_local_torch_device()
+        device = get_torch_device()
         num_inference_steps = batch.num_inference_steps
         timesteps = batch.timesteps
         sigmas = batch.sigmas

fastvideo/v1/pipelines/stepvideo/stepvideo_pipeline.py

@@ -14,7 +14,7 @@ from typing import Any, Dict
 import torch
 from huggingface_hub import hf_hub_download
 
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device
 from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.models.encoders.bert import HunyuanClip  # type: ignore
@@ -78,7 +78,7 @@ class StepVideoPipeline(LoRAPipeline, ComposedPipelineBase):
         """
         Initialize the pipeline.
         """
-        target_device = get_local_torch_device()
+        target_device = get_torch_device()
         llm_dir = os.path.join(self.model_path, "step_llm")
         clip_dir = os.path.join(self.model_path, "hunyuan_clip")
         text_enc = self.build_llm(llm_dir, target_device)

fastvideo/v1/tests/encoders/test_clip_encoder.py

@@ -6,7 +6,7 @@ import numpy as np
 import pytest
 import torch
 from transformers import AutoConfig
-import gc
+
 from fastvideo.models.hunyuan.text_encoder import (load_text_encoder,
                                                    load_tokenizer)
 # from fastvideo.v1.models.hunyuan.text_encoder import load_text_encoder, load_tokenizer
@@ -16,8 +16,6 @@ from fastvideo.v1.fastvideo_args import FastVideoArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.utils import maybe_download_model
 from fastvideo.v1.configs.models.encoders import CLIPTextConfig
-from torch.distributed.tensor import DTensor
-from torch.testing import assert_close
 
 logger = init_logger(__name__)
 
@@ -68,6 +66,7 @@ def test_clip_encoder():
     # Load the HuggingFace implementation directly
     # model2 = CLIPTextModel(hf_config)
     # model2 = model2.to(torch.float16)
+    model2 = model2.to(device)
     model2.eval()
 
     # Sanity check weights between the two models
@@ -79,20 +78,19 @@ def test_clip_encoder():
     logger.info("Model1 has %d parameters", len(params1))
     logger.info("Model2 has %d parameters", len(params2))
 
-    for name1, param1 in sorted(params1.items()):
-        name2 = name1
-        skip = False
-        for param_name, weight_name, shard_id in model2.config.arch_config.stacked_params_mapping:
-            if weight_name not in name1:
-                skip = True
-        # stacked params are more troublesome
-        if skip:
-            continue
-        param2 = params2[name2]
-        param2 = param2.to_local().to(device) if isinstance(param2, DTensor) else param2.to(device)
-        assert_close(param1, param2, atol=1e-4, rtol=1e-4)
-    gc.collect()
-    torch.cuda.empty_cache()
+    # Compare a few key parameters
+
+    # weight_diffs = []
+    # for (name1, param1), (name2, param2) in zip(
+    #     sorted(params1.items()), sorted(params2.items())
+    # ):
+    #     # if len(weight_diffs) < 5:  # Just check a few parameters
+    #     max_diff = torch.max(torch.abs(param1 - param2)).item()
+    #     mean_diff = torch.mean(torch.abs(param1 - param2)).item()
+    #     weight_diffs.append((name1, name2, max_diff, mean_diff))
+    #     logger.info(f"Parameter: {name1} vs {name2}")
+    #     logger.info(f"  Max diff: {max_diff}, Mean diff: {mean_diff}")
+
     # Load tokenizer
     tokenizer, _ = load_tokenizer(tokenizer_type="clipL",
                                   tokenizer_path=args.model_path,

fastvideo/v1/tests/encoders/test_llama_encoder.py

@@ -5,7 +5,7 @@ import numpy as np
 import pytest
 import torch
 from transformers import AutoConfig
-import gc
+
 from fastvideo.models.hunyuan.text_encoder import (load_text_encoder,
                                                    load_tokenizer)
 from fastvideo.v1.configs.pipelines import PipelineConfig
@@ -15,8 +15,7 @@ from fastvideo.v1.logger import init_logger
 from fastvideo.v1.models.loader.component_loader import TextEncoderLoader
 from fastvideo.v1.utils import maybe_download_model
 from fastvideo.v1.configs.models.encoders import LlamaConfig
-from torch.distributed.tensor import DTensor
-from torch.testing import assert_close
+
 logger = init_logger(__name__)
 
 os.environ["MASTER_ADDR"] = "localhost"
@@ -63,6 +62,7 @@ def test_llama_encoder():
 
     # Convert to float16 and move to device
     # model2 = model2.to(torch.float16)
+    model2 = model2.to(device)
     model2.eval()
 
     # Sanity check weights between the two models
@@ -77,28 +77,34 @@ def test_llama_encoder():
     # Compare a few key parameters
     weight_diffs = []
     # check if embed_tokens are the same
-    device = model1.embed_tokens.weight.device
+    print(model1.embed_tokens.weight.shape, model2.embed_tokens.weight.shape)
     assert torch.allclose(model1.embed_tokens.weight,
-                          model2.embed_tokens.weight.to_local().to(device) if isinstance(model2.embed_tokens.weight, DTensor) else model2.embed_tokens.weight.to(device))
+                          model2.embed_tokens.weight)
     weights = [
         "layers.{}.input_layernorm.weight",
         "layers.{}.post_attention_layernorm.weight"
     ]
-
-    for name1, param1 in sorted(params1.items()):
-        name2 = name1
-        skip = False
-        for param_name, weight_name, shard_id in model2.config.arch_config.stacked_params_mapping:
-            if weight_name not in name1:
-                skip = True
-        # stacked params are more troublesome
-        if skip:
-            continue
-        param2 = params2[name2]
-        param2 = param2.to_local().to(device) if isinstance(param2, DTensor) else param2.to(device)
-        assert_close(param1, param2, atol=1e-4, rtol=1e-4)
-    gc.collect()
-    torch.cuda.empty_cache()
+    # for (name1, param1), (name2, param2) in zip(
+    #     sorted(params1.items()), sorted(params2.items())
+    # ):
+    for layer_idx in range(hf_config.num_hidden_layers):
+        for w in weights:
+            name1 = w.format(layer_idx)
+            name2 = w.format(layer_idx)
+            p1 = params1[name1]
+            p2 = params2[name2]
+            # print(type(p2))
+            if "gate_up" in name2:
+                # print("skipping gate_up")
+                continue
+            try:
+                # logger.info(f"Parameter: {name1} vs {name2}")
+                max_diff = torch.max(torch.abs(p1 - p2)).item()
+                mean_diff = torch.mean(torch.abs(p1 - p2)).item()
+                weight_diffs.append((name1, name2, max_diff, mean_diff))
+                # logger.info(f"  Max diff: {max_diff}, Mean diff: {mean_diff}")
+            except Exception as e:
+                logger.info("Error comparing %s and %s: %s", name1, name2, e)
 
     tokenizer, _ = load_tokenizer(tokenizer_type="llm",
                                   tokenizer_path=TOKENIZER_PATH,

fastvideo/v1/tests/encoders/test_t5_encoder.py

@@ -4,8 +4,6 @@ import os
 import numpy as np
 import pytest
 import torch
-from torch.distributed.tensor import DTensor
-from torch.testing import assert_close
 from transformers import AutoConfig, AutoTokenizer, UMT5EncoderModel
 
 from fastvideo.v1.configs.pipelines import PipelineConfig
@@ -43,13 +41,13 @@ def test_t5_encoder():
     tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_PATH)
 
 
-    args = FastVideoArgs(model_path=TEXT_ENCODER_PATH,
-                        pipeline_config=PipelineConfig(text_encoder_configs=(T5Config(),),
-                        text_encoder_precisions=(precision_str,)),
-                        pin_cpu_memory=False)
+    args = FastVideoArgs(model_path=TEXT_ENCODER_PATH, pipeline_config=PipelineConfig(text_encoder_configs=(T5Config(),), text_encoder_precisions=(precision_str,)))
     loader = TextEncoderLoader()
     model2 = loader.load(TEXT_ENCODER_PATH, "", args)
-    model2 = model2.to(precision)
+
+    # Convert to float16 and move to device
+    # model2 = model2.to(precision)
+    model2 = model2.to(device)
     model2.eval()
 
     # Sanity check weights between the two models
@@ -66,17 +64,23 @@ def test_t5_encoder():
     weights = ["encoder.block.{}.layer.0.layer_norm.weight", "encoder.block.{}.layer.0.SelfAttention.relative_attention_bias.weight", \
                "encoder.block.{}.layer.0.SelfAttention.o.weight", "encoder.block.{}.layer.1.DenseReluDense.wi_0.weight", "encoder.block.{}.layer.1.DenseReluDense.wi_1.weight",\
                 "encoder.block.{}.layer.1.DenseReluDense.wo.weight", \
-                "encoder.block.{}.layer.1.layer_norm.weight", "encoder.final_layer_norm.weight"]
-    
+                "encoder.block.{}.layer.1.layer_norm.weight", "encoder.final_layer_norm.weight", "shared.weight"]
     for idx in range(hf_config.num_hidden_layers):
         for w in weights:
             name1 = w.format(idx)
             name2 = w.format(idx)
             p1 = params1[name1]
             p2 = params2[name2]
-            p2 = (p2.to_local() if isinstance(p2, DTensor) else p2).to(p1)
-            assert_close(p1, p2, atol=1e-4, rtol=1e-4)
-    
+            assert p1.dtype == p2.dtype
+            try:
+                logger.info("Parameter: %s vs %s", name1, name2)
+                max_diff = torch.max(torch.abs(p1 - p2)).item()
+                mean_diff = torch.mean(torch.abs(p1 - p2)).item()
+                weight_diffs.append((name1, name2, max_diff, mean_diff))
+                logger.info("  Max diff: %s, Mean diff: %s", max_diff,
+                            mean_diff)
+            except Exception as e:
+                logger.info("Error comparing %s and %s: %s", name1, name2, e)
 
     # Test with some sample prompts
     prompts = [

fastvideo/v1/tests/inference/STA/test_STA_inference.py

@@ -5,7 +5,7 @@ from pathlib import Path
 import pytest
 
 NUM_NODES = "1"
-NUM_GPUS_PER_NODE = "1"
+NUM_GPUS_PER_NODE = "2"
 
 # Set environment variables
 os.environ["FASTVIDEO_ATTENTION_CONFIG"] = "assets/mask_strategy_wan.json"
@@ -17,9 +17,9 @@ def test_inference():
     cmd = [
         "fastvideo", "generate",
         "--model-path", "Wan-AI/Wan2.1-T2V-14B-Diffusers",
-        "--sp-size", "1",
-        "--tp-size", "1",
-        "--num-gpus", "1",
+        "--sp-size", "2",
+        "--tp-size", "2",
+        "--num-gpus", "2",
         "--height", "768",
         "--width", "1280",
         "--num-frames", "69",

fastvideo/v1/tests/modal/pr_test.py

@@ -69,11 +69,11 @@ def run_ssim_tests():
 def run_training_tests():
     run_test("wandb login $WANDB_API_KEY && pytest ./fastvideo/v1/tests/training/Vanilla -srP")
 
-@app.function(gpu="H100:1", image=image, timeout=1800, secrets=[modal.Secret.from_dict({"WANDB_API_KEY": os.environ.get("WANDB_API_KEY", "")})])
+@app.function(gpu="H100:2", image=image, timeout=1800, secrets=[modal.Secret.from_dict({"WANDB_API_KEY": os.environ.get("WANDB_API_KEY", "")})])
 def run_training_tests_VSA():
     run_test("wandb login $WANDB_API_KEY && pytest ./fastvideo/v1/tests/training/VSA -srP")
 
-@app.function(gpu="H100:1", image=image, timeout=1800)
+@app.function(gpu="H100:2", image=image, timeout=1800)
 def run_inference_tests_STA():
     run_test("pytest ./fastvideo/v1/tests/inference/STA -srP")
 

fastvideo/v1/tests/training/VSA/reference_wandb_summary_VSA.json

@@ -1 +1 @@
-{"grad_norm":0.478515625,"_runtime":95.727033597,"_wandb":{"runtime":95},"_step":5,"validation_videos_50_steps":{"videos":[{"_type":"video-file","sha256":"42a1c311521a9d460db788713be1cbf2db767494e02619b43be5bf3eed8381d8","size":158632,"path":"media/videos/validation_videos_50_steps_0_42a1c311521a9d460db7.mp4"},{"path":"media/videos/validation_videos_50_steps_0_818505095b4b5e8b7f51.mp4","_type":"video-file","sha256":"818505095b4b5e8b7f511012d45f04d151ce3344bc058fc0f3225a414a851e4a","size":147825},{"sha256":"fc334ba9ed5e66c8527ee3b408e3be2d76167fef03588bf2840f4a0792f2fe34","size":136933,"path":"media/videos/validation_videos_50_steps_0_fc334ba9ed5e66c8527e.mp4","_type":"video-file"},{"size":201797,"path":"media/videos/validation_videos_50_steps_0_ccd98f6f907635d266a7.mp4","_type":"video-file","sha256":"ccd98f6f907635d266a74783688e7ecf1dac752d79d72d69eab9ef0e3f7413eb"},{"_type":"video-file","sha256":"ca79f40a0aed38f676f12779b349ce40e9e3fb7f36c578f49a20854c70508fb4","size":147114,"path":"media/videos/validation_videos_50_steps_0_ca79f40a0aed38f676f1.mp4"},{"size":175104,"path":"media/videos/validation_videos_50_steps_0_32c9b33ff920c17e5881.mp4","_type":"video-file","sha256":"32c9b33ff920c17e588133d7a27aa400ff3dc529b01ed4f16ac4d6bb2afa0f00"},{"sha256":"2cf520bfb93401c914e93c87ef791c2f12a4e043b95dfdc98115c930e11dfe67","size":139655,"path":"media/videos/validation_videos_50_steps_0_2cf520bfb93401c914e9.mp4","_type":"video-file"},{"_type":"video-file","sha256":"1d73aba17ce582c7aef4af4d64079e3e9d3df205634eff453446bdaf2340b214","size":149028,"path":"media/videos/validation_videos_50_steps_0_1d73aba17ce582c7aef4.mp4"}],"captions":false,"_type":"videos","count":8},"train_loss":0.08922439813613892,"_timestamp":1.750202051751466e+09,"avg_step_time":0.7536672964692116,"step_time":0.4742048177868128,"learning_rate":1e-05,"vsa_sparsity":0.05}
+{"step_time":0.6983645600266755,"_wandb":{"runtime":107},"grad_norm":0.50390625,"avg_step_time":1.002151239803061,"_step":5,"validation_videos_50_steps":{"captions":false,"_type":"videos","count":8,"videos":[{"size":159131,"path":"media/videos/validation_videos_50_steps_0_dc447599dbe48350e9c9.mp4","_type":"video-file","sha256":"dc447599dbe48350e9c920f4971e1786bde580dd20d52b4aa147ae8d3dc564d6"},{"_type":"video-file","sha256":"4e283876ddfbf5a2cb6f8aca07a39f832b5f806fbefde8854bc73ed904ff20ee","size":160315,"path":"media/videos/validation_videos_50_steps_0_4e283876ddfbf5a2cb6f.mp4"},{"size":135225,"path":"media/videos/validation_videos_50_steps_0_78185c41e1935306e93c.mp4","_type":"video-file","sha256":"78185c41e1935306e93c2d416ee40b31d038abffb029cb5bfb11c2a634eb2fcf"},{"_type":"video-file","sha256":"27e9819d002d3f63c8918bbdc5bf2857b5effe0caf5d5b8374b9c590fc6432eb","size":197873,"path":"media/videos/validation_videos_50_steps_0_27e9819d002d3f63c891.mp4"},{"_type":"video-file","sha256":"46fe548e86144ca60a9396fcd15e8788d3a05c066c6819ccdd6a9041feaaec8f","size":170601,"path":"media/videos/validation_videos_50_steps_0_46fe548e86144ca60a93.mp4"},{"sha256":"91ec338774bec870b9c5c81be4330a8f0f2535124cb372e881c3703e6b65ed77","size":164462,"path":"media/videos/validation_videos_50_steps_0_91ec338774bec870b9c5.mp4","_type":"video-file"},{"_type":"video-file","sha256":"ee4e811080a619215fd7541b39203dfb69c2db4cdadbfdd7a234a2cff684f6f3","size":139435,"path":"media/videos/validation_videos_50_steps_0_ee4e811080a619215fd7.mp4"},{"sha256":"22e31e048ba5e5b9d6587d7306904d6edc6c618b8f77c3ceb05ba2d602309274","size":147072,"path":"media/videos/validation_videos_50_steps_0_22e31e048ba5e5b9d658.mp4","_type":"video-file"}]},"_timestamp":1.75118195270901e+09,"vsa_sparsity":0.05,"learning_rate":1e-05,"train_loss":0.19960195198655128,"_runtime":107.325113071}

fastvideo/v1/tests/training/VSA/test_training_loss_VSA.py

@@ -15,7 +15,7 @@ wandb_name = "test_training_loss_VSA"
 reference_wandb_summary_file = "fastvideo/v1/tests/training/VSA/reference_wandb_summary_VSA.json"
 
 NUM_NODES = "1"
-NUM_GPUS_PER_NODE = "1"
+NUM_GPUS_PER_NODE = "2"
 
 os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "VIDEO_SPARSE_ATTN"
 
@@ -35,14 +35,14 @@ def run_worker():
         "--validation_preprocessed_path", "data/mini_dataset_i2v_VSA/validation_parquet_dataset",
         "--train_batch_size", "1",
         "--num_latent_t", "4",
-        "--num_gpus", "1",
-        "--sp_size", "1",
-        "--tp_size", "1",
+        "--num_gpus", "2",
+        "--sp_size", "2",
+        "--tp_size", "2",
         "--hsdp_replicate_dim", "1",
-        "--hsdp_shard_dim", "1",
+        "--hsdp_shard_dim", "2",
         "--train_sp_batch_size", "1",
         "--dataloader_num_workers", "4",
-        "--gradient_accumulation_steps", "1",
+        "--gradient_accumulation_steps", "2",
         "--max_train_steps", "5",
         "--learning_rate", "1e-5",
         "--mixed_precision", "bf16",
@@ -110,7 +110,7 @@ def test_distributed_training():
     fields_and_thresholds = {
         'avg_step_time': 1.0,
         'grad_norm': 0.1,
-        'step_time': 0.5,
+        'step_time': 1.0,
         'train_loss': 0.001
     }
 

fastvideo/v1/tests/training/Vanilla/test_training_loss.py

@@ -121,9 +121,9 @@ def test_distributed_training():
     wandb_summary = json.load(open(summary_file))
 
     fields_and_thresholds = {
-        'avg_step_time': 1.0,
+        'avg_step_time': 6.0,
         'grad_norm': 0.3,
-        'step_time': 0.5,
+        'step_time': 6.0,
         'train_loss': 0.0025
     }
 

fastvideo/v1/tests/transformers/test_hunyuanvideo.py

@@ -80,10 +80,7 @@ def test_hunyuanvideo_distributed():
 
     # Initialize with identical weights
     model = initialize_identical_weights(model, seed=42)
-    shard_model(model, cpu_offload=True,
-                reshard_after_forward=True,
-                fsdp_shard_conditions=model._fsdp_shard_conditions
-                )
+    shard_model(model, cpu_offload=False, reshard_after_forward=True)
     for n, p in chain(model.named_parameters(), model.named_buffers()):
         if p.is_meta:
             raise RuntimeError(

fastvideo/v1/training/training_pipeline.py

@@ -24,9 +24,8 @@ from fastvideo.v1.configs.sample import SamplingParam
 from fastvideo.v1.dataset import build_parquet_map_style_dataloader
 from fastvideo.v1.dataset.dataloader.schema import (
     pyarrow_schema_t2v, pyarrow_schema_t2v_validation)
-from fastvideo.v1.distributed import (cleanup_dist_env_and_memory,
-                                      get_local_torch_device, get_sp_group,
-                                      get_world_group)
+from fastvideo.v1.distributed import (cleanup_dist_env_and_memory, get_sp_group,
+                                      get_torch_device, get_world_group)
 from fastvideo.v1.fastvideo_args import FastVideoArgs, TrainingArgs
 from fastvideo.v1.forward_context import set_forward_context
 from fastvideo.v1.logger import init_logger
@@ -66,8 +65,8 @@ class TrainingPipeline(ComposedPipelineBase, ABC):
 
     def initialize_training_pipeline(self, training_args: TrainingArgs):
         logger.info("Initializing training pipeline...")
-        self.device = get_local_torch_device()
         self.training_args = training_args
+        self.device = get_torch_device()
         world_group = get_world_group()
         self.world_size = world_group.world_size
         self.global_rank = world_group.rank
@@ -177,12 +176,12 @@ class TrainingPipeline(ComposedPipelineBase, ABC):
         encoder_attention_mask = batch['text_attention_mask']
         infos = batch['info_list']
 
-        training_batch.latents = latents.to(get_local_torch_device(),
+        training_batch.latents = latents.to(get_torch_device(),
                                             dtype=torch.bfloat16)
         training_batch.encoder_hidden_states = encoder_hidden_states.to(
-            get_local_torch_device(), dtype=torch.bfloat16)
+            get_torch_device(), dtype=torch.bfloat16)
         training_batch.encoder_attention_mask = encoder_attention_mask.to(
-            get_local_torch_device(), dtype=torch.bfloat16)
+            get_torch_device(), dtype=torch.bfloat16)
         training_batch.infos = infos
 
         return training_batch
@@ -246,9 +245,8 @@ class TrainingPipeline(ComposedPipelineBase, ABC):
         current_vsa_sparsity = training_batch.current_vsa_sparsity
 
         if vsa_available and envs.FASTVIDEO_ATTENTION_BACKEND == "VIDEO_SPARSE_ATTN":
-
             dit_seq_shape = [
-                latents.shape[2] // patch_size[0],
+                latents.shape[2] * self.sp_world_size // patch_size[0],
                 latents.shape[3] // patch_size[1],
                 latents.shape[4] // patch_size[2]
             ]
@@ -275,7 +273,7 @@ class TrainingPipeline(ComposedPipelineBase, ABC):
             "encoder_hidden_states":
             training_batch.encoder_hidden_states,
             "timestep":
-            training_batch.timesteps.to(get_local_torch_device(),
+            training_batch.timesteps.to(get_torch_device(),
                                         dtype=torch.bfloat16),
             "encoder_attention_mask":
             training_batch.encoder_attention_mask,
@@ -552,9 +550,8 @@ class TrainingPipeline(ComposedPipelineBase, ABC):
         prompt_embeds = validation_batch['text_embedding']
         prompt_attention_mask = validation_batch['text_attention_mask']
 
-        prompt_embeds = prompt_embeds.to(get_local_torch_device())
-        prompt_attention_mask = prompt_attention_mask.to(
-            get_local_torch_device())
+        prompt_embeds = prompt_embeds.to(get_torch_device())
+        prompt_attention_mask = prompt_attention_mask.to(get_torch_device())
 
         # Calculate sizes
         latents_size = [(sampling_param.num_frames - 1) // 4 + 1,

fastvideo/v1/training/wan_i2v_training_pipeline.py

@@ -4,11 +4,12 @@ from copy import deepcopy
 from typing import Any, Dict
 
 import torch
+import torch.distributed
 
 from fastvideo.v1.configs.sample import SamplingParam
 from fastvideo.v1.dataset.dataloader.schema import (
     pyarrow_schema_i2v, pyarrow_schema_i2v_validation)
-from fastvideo.v1.distributed import get_local_torch_device
+from fastvideo.v1.distributed import get_torch_device, get_sp_group
 from fastvideo.v1.fastvideo_args import FastVideoArgs, TrainingArgs
 from fastvideo.v1.logger import init_logger
 from fastvideo.v1.models.schedulers.scheduling_flow_unipc_multistep import (
@@ -18,6 +19,8 @@ from fastvideo.v1.pipelines.pipeline_batch_info import (ForwardBatch,
 from fastvideo.v1.pipelines.wan.wan_i2v_pipeline import (
     WanImageToVideoValidationPipeline)
 from fastvideo.v1.training.training_pipeline import TrainingPipeline
+from fastvideo.v1.training.training_utils import (shard_latents_across_sp, 
+                                                   clip_grad_norm_while_handling_failing_dtensor_cases)
 from fastvideo.v1.utils import is_vsa_available
 
 vsa_available = is_vsa_available()
@@ -85,17 +88,15 @@ class WanI2VTrainingPipeline(TrainingPipeline):
         pil_image = batch['pil_image']
         infos = batch['info_list']
 
-        training_batch.latents = latents.to(get_local_torch_device(),
+        training_batch.latents = latents.to(get_torch_device(),
                                             dtype=torch.bfloat16)
         training_batch.encoder_hidden_states = encoder_hidden_states.to(
-            get_local_torch_device(), dtype=torch.bfloat16)
+            get_torch_device(), dtype=torch.bfloat16)
         training_batch.encoder_attention_mask = encoder_attention_mask.to(
-            get_local_torch_device(), dtype=torch.bfloat16)
-        training_batch.preprocessed_image = pil_image.to(
-            get_local_torch_device())
-        training_batch.image_embeds = clip_features.to(get_local_torch_device())
-        training_batch.image_latents = image_latents.to(
-            get_local_torch_device())
+            get_torch_device(), dtype=torch.bfloat16)
+        training_batch.preprocessed_image = pil_image.to(get_torch_device())
+        training_batch.image_embeds = clip_features.to(get_torch_device())
+        training_batch.image_latents = image_latents.to(get_torch_device())
         training_batch.infos = infos
 
         return training_batch
@@ -114,8 +115,8 @@ class WanI2VTrainingPipeline(TrainingPipeline):
         training_batch = super()._prepare_dit_inputs(training_batch)
 
         assert isinstance(training_batch.image_latents, torch.Tensor)
-        image_latents = training_batch.image_latents.to(
-            get_local_torch_device(), dtype=torch.bfloat16)
+        image_latents = training_batch.image_latents.to(get_torch_device(),
+                                                        dtype=torch.bfloat16)
 
         training_batch.noisy_model_input = torch.cat(
             [training_batch.noisy_model_input, image_latents], dim=1)
@@ -134,8 +135,7 @@ class WanI2VTrainingPipeline(TrainingPipeline):
         # Image Embeds for conditioning
         image_embeds = training_batch.image_embeds
         assert torch.isnan(image_embeds).sum() == 0
-        image_embeds = image_embeds.to(get_local_torch_device(),
-                                       dtype=torch.bfloat16)
+        image_embeds = image_embeds.to(get_torch_device(), dtype=torch.bfloat16)
         encoder_hidden_states_image = image_embeds
 
         # NOTE: noisy_model_input already contains concatenated image_latents from _prepare_dit_inputs
@@ -145,7 +145,7 @@ class WanI2VTrainingPipeline(TrainingPipeline):
             "encoder_hidden_states":
             training_batch.encoder_hidden_states,
             "timestep":
-            training_batch.timesteps.to(get_local_torch_device(),
+            training_batch.timesteps.to(get_torch_device(),
                                         dtype=torch.bfloat16),
             "encoder_attention_mask":
             training_batch.encoder_attention_mask,
@@ -169,9 +169,9 @@ class WanI2VTrainingPipeline(TrainingPipeline):
         infos = validation_batch['info_list']
         prompt = infos[0]['prompt']
 
-        prompt_embeds = embeddings.to(get_local_torch_device())
-        prompt_attention_mask = masks.to(get_local_torch_device())
-        clip_features = clip_features.to(get_local_torch_device())
+        prompt_embeds = embeddings.to(get_torch_device())
+        prompt_attention_mask = masks.to(get_torch_device())
+        clip_features = clip_features.to(get_torch_device())
 
         # Calculate sizes
         latents_size = [(sampling_param.num_frames - 1) // 4 + 1,
@@ -207,6 +207,36 @@ class WanI2VTrainingPipeline(TrainingPipeline):
         )
         return batch
 
+    def _clip_grad_norm(self, training_batch: TrainingBatch) -> TrainingBatch:
+        """Override to add gradient synchronization across SP ranks."""
+        assert self.training_args is not None
+        max_grad_norm = self.training_args.max_grad_norm
+
+        # CRITICAL FIX: Synchronize gradients across SP ranks before clipping
+        # Different SP ranks compute different gradients due to different noise patterns
+        # These gradients must be averaged across SP ranks for stable training
+        if self.training_args.sp_size > 1:
+            sp_group = get_sp_group()
+            for param in self.transformer.parameters():
+                if param.grad is not None:
+                    # Average gradients across SP ranks
+                    sp_group.all_reduce(param.grad, op=torch.distributed.ReduceOp.AVG)
+
+        if max_grad_norm is not None:
+            model_parts = [self.transformer]
+            grad_norm = clip_grad_norm_while_handling_failing_dtensor_cases(
+                [p for m in model_parts for p in m.parameters()],
+                max_grad_norm,
+                foreach=None,
+            )
+            assert grad_norm is not float('nan') or grad_norm is not float(
+                'inf')
+            grad_norm = grad_norm.item() if grad_norm is not None else 0.0
+        else:
+            grad_norm = 0.0
+        training_batch.grad_norm = grad_norm
+        return training_batch
+
 
 def main(args) -> None:
     logger.info("Starting training pipeline...")