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, 23:15 UTC
Add basic correctness 2 GPU tests to 4 GPU pipeline (#5518)
Antoni Baum committed
Jun 16, 2024 at 00:48 UTC
f31c1f90e381967d25591a8928782d8a6a13693e
1 file changed
+6
-2
.buildkite/test-pipeline.yaml
+6
-2
@@ -32,7 +32,7 @@ steps:
32
working_dir: "/vllm-workspace/tests"
33
num_gpus: 2
34
35
-- label: Distributed Tests
35
+- label: Distributed Tests (2 GPUs)
36
mirror_hardwares: [amd]
37
working_dir: "/vllm-workspace/tests"
38
num_gpus: 2
@@ -50,12 +50,16 @@ steps:
50
- CUDA_VISIBLE_DEVICES=0,1 pytest -v -s test_sharded_state_loader.py
51
- CUDA_VISIBLE_DEVICES=0,1 pytest -v -s distributed/test_utils.py
52
53
-- label: Distributed Tests (Multiple Groups)
53
+- label: Distributed Tests (4 GPUs)
54
#mirror_hardwares: [amd]
55
working_dir: "/vllm-workspace/tests"
56
num_gpus: 4
57
commands:
58
- pytest -v -s distributed/test_pynccl.py
59
+ # We want to test that models which use 2 GPUs work with 4 GPUs, which is why we duplicate them here.
60
+ # See https://github.com/vllm-project/vllm/pull/5473#issuecomment-2166601837 for context.
61
+ - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_basic_distributed_correctness.py
62
+ - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_basic_distributed_correctness.py
63
64
- label: Engine Test
65
mirror_hardwares: [amd]