biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

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

Browse Source 
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 06:46:42 +02:00
committed by GitHub
parent 90805ff464
commit dea268336f
53 changed files with 3613 additions and 1016 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
tests/models/test_vision.py
View File

@@ -6,7 +6,7 @@ import pytest
import torch
import torch.multiprocessing as mp
from tests.utils import multi_gpu_test
from tests.utils import ensure_current_vllm_config, multi_gpu_test
from vllm.distributed import get_tensor_model_parallel_world_size
from vllm.distributed.parallel_state import (
init_distributed_environment,
@@ -117,7 +117,8 @@ def run_dp_sharded_vision_model_vs_direct(
# initialize distributed
init_distributed_environment()
initialize_model_parallel(tensor_model_parallel_size=world_size)
with ensure_current_vllm_config():
initialize_model_parallel(tensor_model_parallel_size=world_size)
# Create a test input tensor
image_input = torch.randn(batch_size, 3, 224, 224)
@@ -302,7 +303,8 @@ def run_dp_sharded_mrope_vision_model_vs_direct(
# initialize distributed
init_distributed_environment()
initialize_model_parallel(tensor_model_parallel_size=world_size)
with ensure_current_vllm_config():
initialize_model_parallel(tensor_model_parallel_size=world_size)
# Create test data
grid_thw_list = []
@@ -377,7 +379,8 @@ def run_dp_sharded_mrope_vision_model_empty_input_worker(
)
init_distributed_environment()
initialize_model_parallel(tensor_model_parallel_size=world_size)
with ensure_current_vllm_config():
initialize_model_parallel(tensor_model_parallel_size=world_size)
# Create empty inputs
pixel_values = torch.empty((0, 768))
@@ -425,7 +428,8 @@ def run_dp_sharded_mrope_vision_model_uneven_load_worker(
)
init_distributed_environment()
initialize_model_parallel(tensor_model_parallel_size=world_size)
with ensure_current_vllm_config():
initialize_model_parallel(tensor_model_parallel_size=world_size)
# Create images with very different sizes
grid_thw_list = [