20231088/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Misc][TPU] Support TPU in initialize_ray_cluster (#6812)

Browse Source
This commit is contained in:
Woosuk Kwon 2024-07-26 12:39:49 -07:00 committed by GitHub
parent 71734f1bf2
commit aa4867791e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 21 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
vllm/executor/ray_utils.py
View File

@ -3,7 +3,7 @@ from typing import List, Optional, Tuple
from vllm.config import ParallelConfig from vllm.config import ParallelConfig
from vllm.logger import init_logger from vllm.logger import init_logger
from vllm.sequence import ExecuteModelRequest from vllm.sequence import ExecuteModelRequest
from vllm.utils import get_ip, is_hip, is_xpu from vllm.utils import get_ip, is_hip, is_tpu, is_xpu
from vllm.worker.worker_base import WorkerWrapperBase from vllm.worker.worker_base import WorkerWrapperBase
logger = init_logger(__name__) logger = init_logger(__name__)
@ -93,32 +93,38 @@ def initialize_ray_cluster(
# Placement group is already set. # Placement group is already set.
return return
device_str = "GPU" if not is_tpu() else "TPU"
# Create placement group for worker processes # Create placement group for worker processes
current_placement_group = ray.util.get_current_placement_group() current_placement_group = ray.util.get_current_placement_group()
if current_placement_group: if current_placement_group:
# We are in a placement group # We are in a placement group
bundles = current_placement_group.bundle_specs bundles = current_placement_group.bundle_specs
# Verify that we can use the placement group. # Verify that we can use the placement group.
gpu_bundles = 0 device_bundles = 0
for bundle in bundles: for bundle in bundles:
bundle_gpus = bundle.get("GPU", 0) bundle_devices = bundle.get(device_str, 0)
if bundle_gpus > 1: if bundle_devices > 1:
raise ValueError( raise ValueError(
"Placement group bundle cannot have more than 1 GPU.") "Placement group bundle cannot have more than 1 "
if bundle_gpus: f"{device_str}.")
gpu_bundles += 1 if bundle_devices:
if parallel_config.world_size > gpu_bundles: device_bundles += 1
if parallel_config.world_size > device_bundles:
raise ValueError( raise ValueError(
"The number of required GPUs exceeds the total number of " f"The number of required {device_str}s exceeds the total "
"available GPUs in the placement group.") f"number of available {device_str}s in the placement group."
f"Required number of devices: {parallel_config.world_size}. "
f"Total number of devices: {device_bundles}.")
else: else:
num_gpus_in_cluster = ray.cluster_resources().get("GPU", 0) num_devices_in_cluster = ray.cluster_resources().get(device_str, 0)
if parallel_config.world_size > num_gpus_in_cluster: if parallel_config.world_size > num_devices_in_cluster:
raise ValueError( raise ValueError(
"The number of required GPUs exceeds the total number of " f"The number of required {device_str}s exceeds the total "
"available GPUs in the cluster.") f"number of available {device_str}s in the placement group.")
# Create a new placement group # Create a new placement group
placement_group_specs = ([{"GPU": 1}] * parallel_config.world_size) placement_group_specs = ([{
device_str: 1
}] * parallel_config.world_size)
current_placement_group = ray.util.placement_group( current_placement_group = ray.util.placement_group(
placement_group_specs) placement_group_specs)
# Wait until PG is ready - this will block until all # Wait until PG is ready - this will block until all