kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-14 06:27:45 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[None][fix] Setup dist for AutoTuner in Layerwise benchmarking. (#10534)

Browse Source 
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
This commit is contained in:
Yukun He 2026-01-09 14:16:39 +08:00 committed by GitHub
parent 6fcd4e7099
commit c5331e6dbb
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
examples/layer_wise_benchmarks/run.py
View File

@ -10,10 +10,11 @@ import torch
import yaml import yaml
from tensorrt_llm._torch.autotuner import AutoTuner, autotune from tensorrt_llm._torch.autotuner import AutoTuner, autotune
from tensorrt_llm._torch.distributed import MPIDist, TorchDist
from tensorrt_llm._torch.modules.fused_moe.fused_moe_cutlass import CutlassFusedMoE from tensorrt_llm._torch.modules.fused_moe.fused_moe_cutlass import CutlassFusedMoE
from tensorrt_llm._torch.modules.fused_moe.interface import AlltoallMethodType from tensorrt_llm._torch.modules.fused_moe.interface import AlltoallMethodType
from tensorrt_llm._torch.modules.multi_stream_utils import with_multi_stream from tensorrt_llm._torch.modules.multi_stream_utils import with_multi_stream
from tensorrt_llm._utils import local_mpi_rank, mpi_rank, mpi_world_size from tensorrt_llm._utils import local_mpi_rank, mpi_disabled, mpi_rank, mpi_world_size
from tensorrt_llm.logger import logger from tensorrt_llm.logger import logger
from tensorrt_llm.tools.layer_wise_benchmarks import BalanceMethod, get_runner_cls, mark_ranges from tensorrt_llm.tools.layer_wise_benchmarks import BalanceMethod, get_runner_cls, mark_ranges
@ -173,6 +174,8 @@ run_pack = runner.create_run_pack(
) )
if args.enable_autotuner: if args.enable_autotuner:
cache_path = os.getenv("TLLM_AUTOTUNER_CACHE_PATH") or None cache_path = os.getenv("TLLM_AUTOTUNER_CACHE_PATH") or None
dist = TorchDist(mapping=mapping) if mpi_disabled() else MPIDist(mapping=mapping)
AutoTuner.get().setup_distributed_state(mapping, dist)
with autotune(cache_path=cache_path): with autotune(cache_path=cache_path):
run_pack() run_pack()
else: else: