[1/N] Elastic EP Milestone 2 (#34861)

Signed-off-by: Yongji Wu <wuyongji317@gmail.com>
Signed-off-by: Itay Alroy <ialroy@nvidia.com>
Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Signed-off-by: Ron Tourgeman <rtourgeman@nvidia.com>
Co-authored-by: Yongji Wu <wuyongji317@gmail.com>
Co-authored-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Co-authored-by: Ron Tourgeman <rtourgeman@nvidia.com>
This commit is contained in:
Itay Alroy
2026-02-28 04:46:42 +00:00
committed by GitHub
co-authored by Yongji Wu Tyler Michael Smith Ron Tourgeman
parent 90805ff464
commit dea268336f
53 changed files with 3612 additions and 1015 deletions
+12 -9
View File
@@ -45,21 +45,24 @@ def cleanup_fixture(should_do_global_cleanup_after_test: bool):
@pytest.fixture
def dist_init():
from tests.utils import ensure_current_vllm_config
temp_file = tempfile.mkstemp()[1]
backend = "nccl"
if current_platform.is_cpu() or current_platform.is_tpu():
backend = "gloo"
init_distributed_environment(
world_size=1,
rank=0,
distributed_init_method=f"file://{temp_file}",
local_rank=0,
backend=backend,
)
initialize_model_parallel(1, 1)
yield
with ensure_current_vllm_config():
init_distributed_environment(
world_size=1,
rank=0,
distributed_init_method=f"file://{temp_file}",
local_rank=0,
backend=backend,
)
initialize_model_parallel(1, 1)
yield
cleanup_dist_env_and_memory(shutdown_ray=True)
+3 -2
View File
@@ -6,7 +6,7 @@ import random
import pytest
import torch
from tests.utils import multi_gpu_test
from tests.utils import ensure_current_vllm_config, multi_gpu_test
from vllm import _custom_ops as ops
from vllm.distributed import (
init_distributed_environment,
@@ -631,7 +631,8 @@ def use_fused_moe_lora_kernel_tensor_parallel(
local_rank=local_rank,
distributed_init_method=init_method,
)
initialize_model_parallel(world_size, 1)
with ensure_current_vllm_config():
initialize_model_parallel(world_size, 1)
tp_size = get_tensor_model_parallel_world_size()
input_dim = K if column_parallel else N
+4 -2
View File
@@ -13,6 +13,7 @@ from vllm.config import (
ParallelConfig,
SchedulerConfig,
VllmConfig,
set_current_vllm_config,
)
from vllm.config.load import LoadConfig
from vllm.config.lora import LoRAConfig
@@ -77,8 +78,9 @@ def test_worker_apply_lora(qwen3_lora_files):
distributed_init_method=f"file://{tempfile.mkstemp()[1]}",
)
worker.init_device()
worker.load_model()
with set_current_vllm_config(vllm_config):
worker.init_device()
worker.load_model()
set_active_loras(worker, [])
assert worker.list_loras() == set()