课程5：AttributeError: module 'tvm.meta_schedule' has no attribute 'TuneConfig'这是什么问题

[JJYGood]

from tvm import meta_schedule as ms

sch_tuned = ms.tune_tir(
mod=MyModule,
target="llvm --num-cores=1",
config=ms.TuneConfig(
max_trials_global=64,
num_trials_per_iter=64,
),
space=ms.space_generator.ScheduleFn(stochastic_schedule_mm),
work_dir="./tune_tmp",
task_name="main"
)

---

AttributeError Traceback (most recent call last)
in
4 mod=MyModule,
5 target="llvm --num-cores=1",
----> 6 config=ms.TuneConfig(
7 max_trials_global=64,
8 num_trials_per_iter=64,

AttributeError: module 'tvm.meta_schedule' has no attribute 'TuneConfig'

[ticoAg]

应该是使用的tvm版本有差异 当前默认装的版本$max_trials_global$和$num_trials_per_iter$直接作为tune_tir的参数
run默认的代码需在colab执行

!wget https://github.com/mlc-ai/utils/releases/download/v0.9.dev0/mlc_ai_nightly_cu110-0.9.dev2227%2Bg06c0ef3be-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
!pip install ./mlc_ai_nightly_cu110-0.9.dev2227+g06c0ef3be-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

[JJYGood]

感谢，之后装了带有cuda版本的是能成功的

…

------------------ 原始邮件 ------------------ 发件人: ***@***.***>; 发送时间: 2022年11月7日(星期一) 上午10:42 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [mlc-ai/mlc-zh] 课程5：AttributeError: module 'tvm.meta_schedule' has no attribute 'TuneConfig'这是什么问题 (Discussion #162) 应该是使用的tvm版本有差异 当前默认装的版本$max_trials_global$和$num_trials_per_iter$直接作为tune_tir的参数 run默认的代码需在colab执行 !wget https://github.com/mlc-ai/utils/releases/download/v0.9.dev0/mlc_ai_nightly_cu110-0.9.dev2227%2Bg06c0ef3be-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl !pip install ./mlc_ai_nightly_cu110-0.9.dev2227+g06c0ef3be-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[wangyeahua]

API变了， 最新API可以参考tvm/python/tvm/meta_schedule/tir_integration.py 的134行tune_tir的定义。
可以修改你的代码如下：
``
db = ms.tune_tir(
mod=MyModule,
target="llvm --num-cores=1",
# config=ms.TuneConfig(
# max_trials_global=64,
# num_trials_per_iter=64,
# ),
max_trials_global=64,
num_trials_per_iter=64,
space=ms.space_generator.ScheduleFn(stochastic_schedule_mm),
work_dir="./tune_tmp",
task_name="main"
)
tuned_sch=ms.tir_integration.compile_tir(db,MyModule,target="llvm")