mod, params = relax.frontend.detach_params(mod)
mod = relax.get_pipeline("zero")(mod)
db = ms.tir_integration.tune_tir(
mod=mod,
target=target,
work_dir="./tuning_spr_logs",
max_trials_global=8000, # SPR은 코어 많아서 더 돌려야 최고 기록 나옴
max_trials_per_task=100,
num_trials_per_iter=64, # 한 번에 더 많은 후보 테스트
builder=ms.builder.LocalBuilder(),
runner=ms.runner.LocalRunner(evaluator_config=ms.runner.EvaluatorConfig(number=3,
repeat=10)),
database=ms.database.JSONDatabase(work_dir="./tuning_spr_logs"),
)
print("튜닝 완료! 최고 성능 기록이 database에 저장됨")
print(db.get_all_tuning_records()[0])
print(len(db.get_all_tuning_records()))
print(db.has_workload(mod=mod))
Result
...
meta_schedule.TuningRecord(0x7c552c0aaa30)
2345
Fasle
None
With the above code, tuning seems to be recorded well, but
print(db.has_workload(mod=mod)) returns false and
print(db.query(mod=mod,target=target)) returns none.
The db is saved correctly, but is this because of incorrect code, or is there a
problem with the version (0.22)?
---
[Visit Topic](https://discuss.tvm.apache.org/t/error-with-tune-tir/18801/1) to
respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.tvm.apache.org/email/unsubscribe/e6a144f486fa511a62663e3e3bc3d491eb9c7abf5ef19c0ed0e26142051e105f).
