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 5, 02:15 UTC
[CI] Add weight transfer tests to CI (#39821)
Signed-off-by: SumanthRH <sumanthrh99@gmail.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
Sumanth R Hegde committed
Apr 16, 2026 at 12:51 UTC
adf9bb3c577aaaad147b1fe7f61a5c6a0bdfb3de
2 files changed
+5
.buildkite/test_areas/distributed.yaml
+2
@@ -196,6 +196,8 @@ steps:
196
- VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_async_new_apis.py
197
- VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 examples/offline_inference/data_parallel.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput
198
- pytest -v -s tests/v1/distributed/test_dbo.py
199
+ - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py
200
+ - pytest -v -s tests/distributed/test_packed_tensor.py
201
202
- label: Distributed Tests (2 GPUs)(B200)
203
device: b200
tests/distributed/test_weight_transfer.py
+3
@@ -41,6 +41,7 @@ def create_mock_parallel_config(
41
config.rank = rank
42
config.world_size = world_size
43
config.data_parallel_rank = dp_rank
44
+ config.data_parallel_index = dp_rank
45
return config
46
47
@@ -283,6 +284,7 @@ def inference_receive_tensor(
284
parallel_config.rank = 0
285
parallel_config.world_size = 1
286
parallel_config.data_parallel_rank = 0
287
+ parallel_config.data_parallel_index = 0
288
289
engine = NCCLWeightTransferEngine(config, parallel_config)
290
@@ -666,6 +668,7 @@ def inference_receive_ipc_tensor(
668
parallel_config.rank = 0
669
parallel_config.world_size = 1
670
parallel_config.data_parallel_rank = 0
671
+ parallel_config.data_parallel_index = 0
672
673
engine = IPCWeightTransferEngine(config, parallel_config)
674