A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 15:45 UTC
[Bugfix] Ray 2.9.x doesn't expose available_resources_per_node (#8767)
Signed-off-by: darthhexx <darthhexx@gmail.com>
David Newman committed
Sep 25, 2024 at 17:52 UTC
3368c3ab36436af1342a3156971412e9efdb6419
1 file changed
+6
-1
vllm/executor/ray_utils.py
+6
-1
@@ -18,9 +18,14 @@ PG_WAIT_TIMEOUT = 1800
18
19
try:
20
import ray
21
- from ray._private.state import available_resources_per_node
21
from ray.util import placement_group_table
22
from ray.util.placement_group import PlacementGroup
23
+ try:
24
+ from ray._private.state import available_resources_per_node
25
+ except ImportError:
26
+ # Ray 2.9.x doesn't expose `available_resources_per_node`
27
+ from ray._private.state import state as _state
28
+ available_resources_per_node = _state._available_resources_per_node
29
30
class RayWorkerWrapper(WorkerWrapperBase):
31
"""Ray wrapper for vllm.worker.Worker, allowing Worker to be