Merge

xdit-project · Dec 17, 2024 · 475668a · 475668a
1 parent 872a765
commit 475668a
Show file tree

Hide file tree

Showing 6 changed files with 4 additions and 15 deletions.
diff --git a/examples/sd_run.sh b/examples/sd_run.sh
@@ -61,7 +61,6 @@ $OUTPUT_ARGS \
 --warmup_steps 1 \
 --prompt "brown dog laying on the ground with a metal bowl in front of him." \
 --world_size 2 \
---rank 0 \
 $CFG_ARGS \
 $PARALLLEL_VAE \
 $COMPILE_FLAG \

diff --git a/setup.py b/setup.py
@@ -39,7 +39,7 @@ def get_cuda_version():
             "imageio",
             "imageio-ffmpeg",
             "optimum-quanto",
-            "flash_attn>=2.6.3",
+            "flash_attn>=2.7.0",  # flash_attn>=2.7.0 with torch>=2.4.0 wraps ops with torch.ops
             "ray"
         ],
         extras_require={

diff --git a/xfuser/config/args.py b/xfuser/config/args.py
@@ -79,7 +79,6 @@ class xFuserArgs:
     # tensor parallel
     tensor_parallel_degree: int = 1
     split_scheme: Optional[str] = "row"
-    rank: int = 0
     world_size: int = 1
     # pipefusion parallel
     pipefusion_parallel_degree: int = 1
@@ -153,12 +152,6 @@ def add_cli_args(parser: FlexibleArgumentParser):
 
         # Parallel arguments
         parallel_group = parser.add_argument_group("Parallel Processing Options")
-        parallel_group.add_argument(
-            "--rank",
-            type=int,
-            default=0,
-            help="Rank of the process.",
-        )
         parallel_group.add_argument(
             "--world_size",
             type=int,
@@ -384,7 +377,6 @@ def create_config(
                 world_size=self.world_size,
             ),
             world_size=self.world_size,
-            rank=self.rank,
         )
 
         fast_attn_config = FastAttnConfig(

diff --git a/xfuser/config/config.py b/xfuser/config/config.py
@@ -194,7 +194,6 @@ class ParallelConfig:
     tp_config: TensorParallelConfig
     distributed_executor_backend: Optional[str] = None
     world_size: int = 1 # FIXME: remove this
-    rank: int = 0
     worker_cls: str = "xfuser.worker.worker.Worker"
 
     def __post_init__(self):

diff --git a/xfuser/executor/gpu_executor.py b/xfuser/executor/gpu_executor.py
@@ -23,8 +23,7 @@ def _init_executor(self):
         self._init_ray_workers()
 
     def _init_ray_workers(self):
-        placement_group = initialize_ray_cluster(
-            self.engine_config.parallel_config,ray_address="0.0.0.0:6379")
+        placement_group = initialize_ray_cluster(self.engine_config.parallel_config)
 
         # create placement group and worker wrapper instance for lazy load worker
         self.workers = []

diff --git a/xfuser/worker/worker_wrappers.py b/xfuser/worker/worker_wrappers.py
@@ -34,9 +34,9 @@ def update_environs(environs: Dict[str, str]):
 
 
 class RayWorkerWrapper(BaseWorkerWrapper):
-    def __init__(self, engine_config: EngineConfig, bundle_id: int) -> None:
+    def __init__(self, engine_config: EngineConfig, rank: int) -> None:
         super().__init__(engine_config.parallel_config.worker_cls)
-        self.init_worker(engine_config.parallel_config, bundle_id)
+        self.init_worker(engine_config.parallel_config, rank)
 
     def get_node_and_gpu_ids(
         self,