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 15, 18:30 UTC
[ci] feat: make the test_torchrun_example run with tp=2, external_dp=2 (#15172)
Signed-off-by: Chi Zhang <zhangchi.usc1992@bytedance.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com>
Chi Zhang committed
Mar 21, 2025 at 00:30 UTC
086b56824c88d4baa1b324c16250d5ddc42b6208
2 files changed
+6
-2
.buildkite/test-pipeline.yaml
+4
-2
@@ -136,6 +136,10 @@ steps:
136
- examples/offline_inference/rlhf_colocate.py
137
- tests/examples/offline_inference/data_parallel.py
138
commands:
139
+ # test with tp=2 and external_dp=2
140
+ - VLLM_USE_V1=0 torchrun --nproc-per-node=4 distributed/test_torchrun_example.py
141
+ - torchrun --nproc-per-node=4 distributed/test_torchrun_example.py
142
+ # test with internal dp
143
- python3 ../examples/offline_inference/data_parallel.py
144
- pytest -v -s distributed/test_utils.py
145
- pytest -v -s compile/test_basic_correctness.py
@@ -512,8 +516,6 @@ steps:
516
- entrypoints/llm/test_collective_rpc.py
517
commands:
518
- pytest -v -s entrypoints/llm/test_collective_rpc.py
515
- - VLLM_USE_V1=1 torchrun --nproc-per-node=2 distributed/test_torchrun_example.py
516
- - torchrun --nproc-per-node=2 distributed/test_torchrun_example.py
519
- pytest -v -s ./compile/test_basic_correctness.py
520
- pytest -v -s ./compile/test_wrapper.py
521
- VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
tests/distributed/test_torchrun_example.py
+2
@@ -9,6 +9,8 @@ import torch.distributed as dist
9
from vllm import LLM, SamplingParams
10
from vllm.distributed.parallel_state import get_world_group
11
12
+dist.init_process_group(backend="gloo")
13
+
14
# Create prompts
15
prompts = [
16
"Hello, my name is",