You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2022/08/22 20:10:35 UTC
[GitHub] [tvm] Icemist opened a new pull request, #12545: Introducing multi_filter into ConfigSpace autotvm.
Icemist opened a new pull request, #12545:
URL: https://github.com/apache/tvm/pull/12545
Using multi_filter allows to set additional restrictions on ConfigEntity indexes used by avtotvm tuning and exclude the unsuitable in advance.
Co-authored-by: Andrey Malyshev <el...@gmail.com>
Co-authored-by: Egor Churaev <eg...@gmail.com>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r961190625
##########
python/tvm/autotvm/task/space.py:
##########
@@ -822,6 +830,155 @@ def valid(self):
"""
return not bool(self.errors)
+ def is_index_filtered(self, i):
+ """checks if the index satisfies the multi_filter"""
+ if self._shared_filter is None:
+ return True
+
+ if self._shared_filter_cash is None:
+ self._make_shared_filter_cash()
+
+ return self._shared_filter_cash[i]
+
+ def multi_filter(self, **kwargs):
+ "Keeps arg named 'filter'as function as a multi_filter"
+ if self._collect:
+ self._shared_filter_cash = None
+ self._filtered_length = 0
+ self._shared_filter = kwargs.get("filter", lambda x: True)
+
+ @property
+ def total_length(self):
+ """returns the count of the number of indexes"""
+ if self._total_length is None:
+ self._total_length = int(np.prod([len(x) for x in self.space_map.values()]))
+ return self._total_length
+
+ @property
+ def filtered_length(self):
+ """returns the count of the number of indexes satisfying the multifilter"""
+ if self._shared_filter is None:
+ return self.total_length
+
+ if self._shared_filter_cash is None:
+ self._make_shared_filter_cash()
+
+ return self._filtered_length
+
+ @property
+ def dims(self):
+ if self._dims is None:
+ self._dims = [len(x) for x in self.space_map.values()]
+ return self._dims
+
+ def filtered_within_range_length(self, s, e):
+ """returns the count of the number of indexes satisfying the multi_filter in the range"""
+ assert 0 <= s <= e <= self.total_length
+
+ if self._shared_filter is None:
+ return e - s
+ if self._shared_filter_cash is None:
+ self._make_shared_filter_cash()
+ return self.shared_filter_cash[s:e].count(True)
Review Comment:
```suggestion
return self._shared_filter_cash[s:e].count(True)
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] echuraev commented on pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
echuraev commented on PR #12545:
URL: https://github.com/apache/tvm/pull/12545#issuecomment-1236674985
Quickly looked into this PR. I don't really like that now we need to call tsk.config_space.filtered_length instead of len(tsk.config_space). I think that len(smth) is more intuitive. But I didn't dive deeply into the details. I'll do it later. I think that @elvin-n can provide more useful and detailed review.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
Icemist commented on PR #12545:
URL: https://github.com/apache/tvm/pull/12545#issuecomment-1231546175
CC @merrymercy @eqy
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] elvin-n commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
elvin-n commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r964733626
##########
python/tvm/autotvm/task/space.py:
##########
@@ -822,6 +830,155 @@ def valid(self):
"""
return not bool(self.errors)
+ def is_index_filtered(self, i):
+ """checks if the index satisfies the multi_filter"""
+ if self._shared_filter is None:
+ return True
+
+ if self._shared_filter_cash is None:
+ self._make_shared_filter_cash()
+
+ return self._shared_filter_cash[i]
+
+ def multi_filter(self, **kwargs):
+ "Keeps arg named 'filter'as function as a multi_filter"
+ if self._collect:
+ self._shared_filter_cash = None
Review Comment:
don't we need to call clear_shared_filter_cash() instead?
##########
python/tvm/autotvm/tuner/ga_tuner.py:
##########
@@ -49,41 +48,29 @@ def __init__(self, task, pop_size=100, elite_num=3, mutation_prob=0.1):
assert elite_num <= pop_size, "The number of elites must be less than population size"
+ # random initialization
+ self.pop_size = min(self.pop_size, task.config_space.filtered_length)
+ self.elite_num = min(self.pop_size, self.elite_num)
+
# space info
self.space = task.config_space
- self.dim_keys = []
- self.dims = []
- for k, v in self.space.space_map.items():
- self.dim_keys.append(k)
- self.dims.append(len(v))
-
- self.visited = set([])
+ self.visited = set(self.space.sample_ints(self.pop_size))
# current generation
- self.genes = []
+ self.genes = [self.space.point2knob(idx) for idx in self.visited]
self.scores = []
self.elites = []
self.elite_scores = []
self.trial_pt = 0
- # random initialization
- self.pop_size = min(self.pop_size, len(self.space))
- self.elite_num = min(self.pop_size, self.elite_num)
- for _ in range(self.pop_size):
- tmp_gene = point2knob(np.random.randint(len(self.space)), self.dims)
- while knob2point(tmp_gene, self.dims) in self.visited:
- tmp_gene = point2knob(np.random.randint(len(self.space)), self.dims)
-
- self.genes.append(tmp_gene)
- self.visited.add(knob2point(tmp_gene, self.dims))
-
def next_batch(self, batch_size):
ret = []
- for _ in range(batch_size):
+ while len(ret) < batch_size and self.trial_pt < self.space.total_length:
Review Comment:
Should not be there filtered_length? As I understood gen should contain already filtered points
##########
python/tvm/autotvm/task/space.py:
##########
@@ -664,14 +668,18 @@ def __init__(self):
# private dict to provide sugar
self.space_map = OrderedDict() # name -> space
self._collect = True
- self._length = None
+ self._total_length = None
+ self._filtered_length = None
+ self._dims = None
self._entity_map = OrderedDict() # name -> entity
self._constraints = []
self.errors = []
self.code_hash = None
self.flop = 0
self.cost = None
self.is_fallback = False
+ self._shared_filter = None
+ self._shared_filter_cash = None
Review Comment:
I see that all initialization of the cache happens only once when it is not defined.
At the same time if we initialize the cache anyhow, call API initializing the `_shared_filter_cash` and add one more filter, in this case the search space will not correspond to the cached data anymore. Need to invalidate _shared_filter_cache when we add not only multi_filter, but on adding of any filter
##########
python/tvm/autotvm/task/space.py:
##########
@@ -838,22 +995,26 @@ def _add_new_transform(self, space_class, name, axes, policy, **kwargs):
return [Axis(space, i) for i in range(space.num_output)]
return [Axis(None, i) for i in range(space_class.get_num_output(axes, policy, **kwargs))]
- def __len__(self):
Review Comment:
there are pros and cons to remove length. And it seems the number of modified scripts show that len(task.space) became quite public api :( it will be quite painful for endusers if we remove this function
let's return filtered_length by it and add clear comments to declaration of filtered_length and total_length when we need to use each
##########
python/tvm/topi/adreno/conv2d_nchw.py:
##########
@@ -268,7 +268,11 @@ def schedule_conv2d_NCHWc_KCRSk(cfg, s, output):
cfg.define_split("tile_rx", rx, num_outputs=2)
cfg.define_knob("auto_unroll_max_step", [0, 512, 1500])
cfg.define_knob("unroll_explicit", [0, 1])
-
+ cfg.multi_filter(
+ filter=lambda entity: 32
Review Comment:
need to add same for adreno direct nhwc convolution
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r968249527
##########
python/tvm/topi/adreno/conv2d_nchw.py:
##########
@@ -268,7 +268,11 @@ def schedule_conv2d_NCHWc_KCRSk(cfg, s, output):
cfg.define_split("tile_rx", rx, num_outputs=2)
cfg.define_knob("auto_unroll_max_step", [0, 512, 1500])
cfg.define_knob("unroll_explicit", [0, 1])
-
+ cfg.multi_filter(
+ filter=lambda entity: 32
Review Comment:
Added.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] TejashShah commented on pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
TejashShah commented on PR #12545:
URL: https://github.com/apache/tvm/pull/12545#issuecomment-1231829261
cc @csullivan
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r967564191
##########
python/tvm/autotvm/task/space.py:
##########
@@ -664,14 +668,18 @@ def __init__(self):
# private dict to provide sugar
self.space_map = OrderedDict() # name -> space
self._collect = True
- self._length = None
+ self._total_length = None
+ self._filtered_length = None
+ self._dims = None
self._entity_map = OrderedDict() # name -> entity
self._constraints = []
self.errors = []
self.code_hash = None
self.flop = 0
self.cost = None
self.is_fallback = False
+ self._shared_filter = None
+ self._shared_filter_cash = None
Review Comment:
Good point. I fixed it and added a test. You can check it after I reupdate this PR.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] tmoreau89 commented on pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
tmoreau89 commented on PR #12545:
URL: https://github.com/apache/tvm/pull/12545#issuecomment-1249688080
CC @tqchen @merrymercy this PR makes some in depth changes to autoTVM, PTAL
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] masahi merged pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
masahi merged PR #12545:
URL: https://github.com/apache/tvm/pull/12545
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r969902913
##########
python/tvm/autotvm/task/space.py:
##########
@@ -822,11 +826,262 @@ def valid(self):
"""
return not bool(self.errors)
+ def is_index_valid(self, index):
+ """Checks if the index satisfies the multi_filter condition
+
+ Parameters
+ ----------
+ index: int
+ index from the range of the space
+
+ Returns
+ -------
+ valid: bool
+ whether the index meets all the constraints
+ """
+ assert 0 <= index < self.range_length
+ if self._shared_filter is None:
+ return True
+ if self._shared_filter_cache is None:
+ self._make_shared_filter_cache()
+ return self._shared_filter_cache[index]
+
+ def multi_filter(self, filter): # pylint: disable=redefined-builtin
+ """Keeps a function as a multi_filter
Review Comment:
Updated the description and added an example to the header and documentation of the multi-filter function
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r967756604
##########
python/tvm/autotvm/tuner/ga_tuner.py:
##########
@@ -49,41 +48,29 @@ def __init__(self, task, pop_size=100, elite_num=3, mutation_prob=0.1):
assert elite_num <= pop_size, "The number of elites must be less than population size"
+ # random initialization
+ self.pop_size = min(self.pop_size, task.config_space.filtered_length)
+ self.elite_num = min(self.pop_size, self.elite_num)
+
# space info
self.space = task.config_space
- self.dim_keys = []
- self.dims = []
- for k, v in self.space.space_map.items():
- self.dim_keys.append(k)
- self.dims.append(len(v))
-
- self.visited = set([])
+ self.visited = set(self.space.sample_ints(self.pop_size))
# current generation
- self.genes = []
+ self.genes = [self.space.point2knob(idx) for idx in self.visited]
self.scores = []
self.elites = []
self.elite_scores = []
self.trial_pt = 0
- # random initialization
- self.pop_size = min(self.pop_size, len(self.space))
- self.elite_num = min(self.pop_size, self.elite_num)
- for _ in range(self.pop_size):
- tmp_gene = point2knob(np.random.randint(len(self.space)), self.dims)
- while knob2point(tmp_gene, self.dims) in self.visited:
- tmp_gene = point2knob(np.random.randint(len(self.space)), self.dims)
-
- self.genes.append(tmp_gene)
- self.visited.add(knob2point(tmp_gene, self.dims))
-
def next_batch(self, batch_size):
ret = []
- for _ in range(batch_size):
+ while len(ret) < batch_size and self.trial_pt < self.space.total_length:
Review Comment:
You're partially right, there should be a filtered_length, under the condition that it actually gets already filtered points. Right now it is.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] TejashShah commented on pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
TejashShah commented on PR #12545:
URL: https://github.com/apache/tvm/pull/12545#issuecomment-1234776261
cc @vinx13 @junrushao @masahi
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] elvin-n commented on pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
elvin-n commented on PR #12545:
URL: https://github.com/apache/tvm/pull/12545#issuecomment-1249652002
@masahi @merrymercy could you please take a look?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] elvin-n commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
elvin-n commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r967793966
##########
python/tvm/topi/adreno/conv2d_nchw.py:
##########
@@ -268,7 +268,11 @@ def schedule_conv2d_NCHWc_KCRSk(cfg, s, output):
cfg.define_split("tile_rx", rx, num_outputs=2)
cfg.define_knob("auto_unroll_max_step", [0, 512, 1500])
cfg.define_knob("unroll_explicit", [0, 1])
-
+ cfg.multi_filter(
+ filter=lambda entity: 32
Review Comment:
good point
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] elvin-n commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
elvin-n commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r969214283
##########
python/tvm/autotvm/task/space.py:
##########
@@ -822,11 +826,262 @@ def valid(self):
"""
return not bool(self.errors)
+ def is_index_valid(self, index):
+ """Checks if the index satisfies the multi_filter condition
+
+ Parameters
+ ----------
+ index: int
+ index from the range of the space
+
+ Returns
+ -------
+ valid: bool
+ whether the index meets all the constraints
+ """
+ assert 0 <= index < self.range_length
+ if self._shared_filter is None:
+ return True
+ if self._shared_filter_cache is None:
+ self._make_shared_filter_cache()
+ return self._shared_filter_cache[index]
+
+ def multi_filter(self, filter): # pylint: disable=redefined-builtin
+ """Keeps a function as a multi_filter
Review Comment:
it would be better to explain what is a difference between existing filter and new introduced multi_filter
The difference is in the fact that filter restricts the only parameter while multi_filter can restrict combination of parameters. Implementation of the filter and multi_filter is different that lead to the introduction of total_length and filtered_length for multi_filter
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r967563947
##########
python/tvm/autotvm/task/space.py:
##########
@@ -822,6 +830,155 @@ def valid(self):
"""
return not bool(self.errors)
+ def is_index_filtered(self, i):
+ """checks if the index satisfies the multi_filter"""
+ if self._shared_filter is None:
+ return True
+
+ if self._shared_filter_cash is None:
+ self._make_shared_filter_cash()
+
+ return self._shared_filter_cash[i]
+
+ def multi_filter(self, **kwargs):
+ "Keeps arg named 'filter'as function as a multi_filter"
+ if self._collect:
+ self._shared_filter_cash = None
Review Comment:
Done.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r967565258
##########
python/tvm/topi/adreno/conv2d_nchw.py:
##########
@@ -268,7 +268,11 @@ def schedule_conv2d_NCHWc_KCRSk(cfg, s, output):
cfg.define_split("tile_rx", rx, num_outputs=2)
cfg.define_knob("auto_unroll_max_step", [0, 512, 1500])
cfg.define_knob("unroll_explicit", [0, 1])
-
+ cfg.multi_filter(
+ filter=lambda entity: 32
Review Comment:
Added into python\tvm\topi\adreno\conv2d_nhwc.py
It might make sense to add it for the next implementations, too?
python\tvm\topi\adreno\depthwise_conv2d_nchw.py
python\tvm\topi\adreno\depthwise_conv2d_nhwc.py
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] Icemist commented on a diff in pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm.
Posted by GitBox <gi...@apache.org>.
Icemist commented on code in PR #12545:
URL: https://github.com/apache/tvm/pull/12545#discussion_r967564284
##########
python/tvm/autotvm/task/space.py:
##########
@@ -838,22 +995,26 @@ def _add_new_transform(self, space_class, name, axes, policy, **kwargs):
return [Axis(space, i) for i in range(space.num_output)]
return [Axis(None, i) for i in range(space_class.get_num_output(axes, policy, **kwargs))]
- def __len__(self):
Review Comment:
Agreed, restored the original one.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] elvin-n commented on pull request #12545: [AutoTVM] Introducing multi_filter into ConfigSpace autotvm
Posted by GitBox <gi...@apache.org>.
elvin-n commented on PR #12545:
URL: https://github.com/apache/tvm/pull/12545#issuecomment-1254824500
can we merge this PR into main?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org