MCPcopy
hub / github.com/NVIDIA/TensorRT-LLM / setup_engine

Method setup_engine

tensorrt_llm/executor/rpc_worker.py:93–99  ·  view source on GitHub ↗
(self)

Source from the content-addressed store, hash-verified

91 # manages threads internally.
92
93 def setup_engine(self):
94 # Force all the ranks to wait here, and start creating the executor simultaneously.
95 # Only call barrier if we have multiple ranks to avoid hanging in single-process tests
96 if mpi_comm().Get_size() > 1:
97 mpi_comm().barrier()
98
99 super().setup_engine()
100
101 def shutdown(self):
102 logger_debug(f"[worker] RpcWorker #{mpi_rank()} is shutting down",

Callers 5

main_taskMethod · 0.95
setup_methodMethod · 0.45
__init__Method · 0.45
setup_engine_remoteMethod · 0.45
__init__Method · 0.45

Calls 2

mpi_commFunction · 0.90
barrierMethod · 0.45

Tested by 2

setup_methodMethod · 0.36
__init__Method · 0.36