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

8
tests/compile/passes/distributed/test_async_tp.py
View File

@@ -316,7 +316,6 @@ def async_tp_pass_on_test_model(
# initialize distributed
init_distributed_environment()
initialize_model_parallel(tensor_model_parallel_size=world_size)
# configure vllm config for SequenceParallelismPass
vllm_config = VllmConfig()
@@ -334,11 +333,10 @@ def async_tp_pass_on_test_model(
model=model_name, trust_remote_code=True, dtype=dtype, seed=42
)
async_tp_pass = AsyncTPPass(vllm_config)
# Set the global vllm_config for TestBackend which calls
# get_current_vllm_config()
with set_current_vllm_config(vllm_config):
initialize_model_parallel(tensor_model_parallel_size=world_size)
async_tp_pass = AsyncTPPass(vllm_config)
backend = TestBackend(async_tp_pass)
assert (

2
tests/compile/passes/distributed/test_fusion_all_reduce.py
View File

@@ -278,7 +278,6 @@ def all_reduce_fusion_pass_on_test_model(
)
init_distributed_environment()
initialize_model_parallel(tensor_model_parallel_size=world_size)
custom_ops = []
if enable_rms_norm_custom_op:
@@ -304,6 +303,7 @@ def all_reduce_fusion_pass_on_test_model(
model=model_name, trust_remote_code=True, dtype=dtype, seed=42
)
with set_current_vllm_config(vllm_config):
initialize_model_parallel(tensor_model_parallel_size=world_size)
all_reduce_fusion_pass = AllReduceFusionPass(vllm_config)
noop_pass = NoOpEliminationPass(vllm_config)
func_pass = FixFunctionalizationPass(vllm_config)

2
tests/compile/passes/distributed/test_sequence_parallelism.py
View File

@@ -242,7 +242,6 @@ def sequence_parallelism_pass_on_test_model(
# initialize distributed
init_distributed_environment()
initialize_model_parallel(tensor_model_parallel_size=world_size)
# configure vllm config for SequenceParallelismPass
custom_ops_list = custom_ops.split(",") if custom_ops else []
@@ -272,6 +271,7 @@ def sequence_parallelism_pass_on_test_model(
)
with set_current_vllm_config(vllm_config):
initialize_model_parallel(tensor_model_parallel_size=world_size)
noop_pass = NoOpEliminationPass(vllm_config)
sequence_parallelism_pass = SequenceParallelismPass(vllm_config)
cleanup_pass = PostCleanupPass(vllm_config)