Restore buffers when wake up from level 2 sleep (vllm-project#16564) (vllm-project#16889)

Signed-off-by: Han <zh950713@gmail.com>
Signed-off-by: Mu Huai <tianbowen.tbw@antgroup.com>

Author: fingertap

vllm/v1/worker/gpu_worker.py

@@ -54,6 +54,9 @@ def __init__(
             from vllm.utils import init_cached_hf_modules
             init_cached_hf_modules()
 
+        # Buffers saved before sleep
+        self._sleep_saved_buffers: dict[str, torch.Tensor] = {}
+
         # Torch profiler. Enabled and configured through env vars:
         # VLLM_TORCH_PROFILER_DIR=/path/to/save/trace
         if envs.VLLM_TORCH_PROFILER_DIR:
@@ -73,6 +76,15 @@ def __init__(
 
     def sleep(self, level: int = 1) -> None:
         free_bytes_before_sleep = torch.cuda.mem_get_info()[0]
+
+        # Save the buffers before level 2 sleep
+        if level == 2:
+            model = self.model_runner.model
+            self._sleep_saved_buffers = {
+                name: buffer.cpu().clone()
+                for name, buffer in model.named_buffers()
+            }
+
         allocator = CuMemAllocator.get_instance()
         allocator.sleep(offload_tags=("weights", ) if level == 1 else tuple())
         free_bytes_after_sleep, total = torch.cuda.mem_get_info()
@@ -88,6 +100,14 @@ def wake_up(self, tags: Optional[list[str]] = None) -> None:
         allocator = CuMemAllocator.get_instance()
         allocator.wake_up(tags)
 
+        # Restore the buffers after level 2 sleep
+        if len(self._sleep_saved_buffers):
+            model = self.model_runner.model
+            for name, buffer in model.named_buffers():
+                if name in self._sleep_saved_buffers:
+                    buffer.data.copy_(self._sleep_saved_buffers[name].data)
+            self._sleep_saved_buffers = {}
+
     def init_device(self):
         if self.device_config.device.type == "cuda":
             # torch.distributed.all_reduce does not free the input tensor until

vllm/worker/worker.py

@@ -95,6 +95,9 @@ def __init__(
         self.gpu_cache: Optional[List[List[torch.Tensor]]] = None
         self._seq_group_metadata_cache: Dict[str, SequenceGroupMetadata] = {}
 
+        # Buffers saved before sleep
+        self._sleep_saved_buffers: Dict[str, torch.Tensor] = {}
+
         # Torch profiler. Enabled and configured through env vars:
         # VLLM_TORCH_PROFILER_DIR=/path/to/save/trace
         if envs.VLLM_TORCH_PROFILER_DIR:
@@ -124,6 +127,15 @@ def stop_profile(self):
 
     def sleep(self, level: int = 1) -> None:
         free_bytes_before_sleep = torch.cuda.mem_get_info()[0]
+
+        # Save the buffers before level 2 sleep
+        if level == 2:
+            model = self.model_runner.model
+            self._sleep_saved_buffers = {
+                name: buffer.cpu().clone()
+                for name, buffer in model.named_buffers()
+            }
+
         allocator = CuMemAllocator.get_instance()
         allocator.sleep(offload_tags=("weights", ) if level == 1 else tuple())
         free_bytes_after_sleep, total = torch.cuda.mem_get_info()
@@ -139,6 +151,14 @@ def wake_up(self, tags: Optional[list[str]] = None) -> None:
         allocator = CuMemAllocator.get_instance()
         allocator.wake_up(tags=tags)
 
+        # Restore the buffers after level 2 sleep
+        if len(self._sleep_saved_buffers):
+            model = self.model_runner.model
+            for name, buffer in model.named_buffers():
+                if name in self._sleep_saved_buffers:
+                    buffer.data.copy_(self._sleep_saved_buffers[name].data)
+            self._sleep_saved_buffers = {}
+
     def init_device(self) -> None:
         if self.device_config.device.type == "cuda":
             # torch.distributed.all_reduce does not free the input tensor until