masahi commented on a change in pull request #4771: [Relay] Added Merge
Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#discussion_r370507137
##
File path: tests/python/relay/test_pass_merge_composite.py
##
@@ -0,0 +1,158 @@
+# Licensed to the Apache
xin71 commented on issue #4368: [tutorial][benchmark] nnvm -> relay
URL: https://github.com/apache/incubator-tvm/pull/4368#issuecomment-578018262
@yzhliu Hi, I am having hard time to reproduce the results of mobile GPU. I
am wondering did you test this PR w/ mobile gpu? Thanks!!!
-
This is an automated email from the ASF dual-hosted git repository.
kevinthesun pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from 2662125 add missing nullptr check (#4773)
add 9bd2c7b [Bugfix][Frontend][TF] Fix incorrect calcula
kevinthesun commented on issue #4518: [Bugfix][Frontend][TF] Fix incorrect
calculations in tf SLICE
URL: https://github.com/apache/incubator-tvm/pull/4518#issuecomment-578011242
Thanks @optima2005 @inadob
This is an automate
kevinthesun merged pull request #4518: [Bugfix][Frontend][TF] Fix incorrect
calculations in tf SLICE
URL: https://github.com/apache/incubator-tvm/pull/4518
This is an automated message from the Apache Git Service.
To respond
masahi commented on issue #4773: [QNN] Add missing nullptr check
URL: https://github.com/apache/incubator-tvm/pull/4773#issuecomment-577992188
thanks @anijain2305 @vinx13
This is an automated message from the Apache Git Servi
This is an automated email from the ASF dual-hosted git repository.
masahi pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from 69d2f9b [TOPI] Remove cpp upsampling and resize op (#4769)
add 2662125 add missing nullptr check (#4773
masahi merged pull request #4773: [QNN] Add missing nullptr check
URL: https://github.com/apache/incubator-tvm/pull/4773
This is an automated message from the Apache Git Service.
To respond to the message, please log on to Gi
masahi commented on a change in pull request #4771: [Relay] Added Merge
Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#discussion_r370470303
##
File path: tests/python/relay/test_pass_merge_composite.py
##
@@ -0,0 +1,158 @@
+# Licensed to the Apache
masahi edited a comment on issue #4771: [Relay] Added Merge Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#issuecomment-577985906
ok reading your original proposal carefully I understand that I can look up
'Composite' attribute to know if a particular function has a p
masahi edited a comment on issue #4771: [Relay] Added Merge Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#issuecomment-577985906
ok reading your original proposal carefully I understand that I can look up
'Composite' attribute to know if a particular function has a p
masahi commented on issue #4771: [Relay] Added Merge Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#issuecomment-577985906
ok reading your original proposal carefully I understand that I can look up
'Composite' attribute to know if a particular function has a pattern
masahi commented on a change in pull request #4771: [Relay] Added Merge
Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#discussion_r370468299
##
File path: tests/python/relay/test_pass_merge_composite.py
##
@@ -0,0 +1,158 @@
+# Licensed to the Apache
masahi edited a comment on issue #4771: [Relay] Added Merge Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#issuecomment-577983004
From your test cases, I don't see why this pass would help remove pattern
matching on codegen side. Since Relay doesn't have `add_sub_mul`
masahi commented on issue #4771: [Relay] Added Merge Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#issuecomment-577983004
From your test cases, I don't see why this pass would help remove pattern
matching on codegen side. Since Relay doesn't have `add_sub_mul` op, I
vinx13 merged pull request #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769
This is an automated message from the Apache Git Service.
To respond to the message, please l
This is an automated email from the ASF dual-hosted git repository.
wuwei pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from 1ae44cf Fix Tensorflow conv3d pad bug, add non-cubic data and kernel
tests (#4772)
add 69d2f9b [TOPI] R
vinx13 merged pull request #4772: Fix Tensorflow conv3d pad bug, add non-cubic
data and kernel tests
URL: https://github.com/apache/incubator-tvm/pull/4772
This is an automated message from the Apache Git Service.
To respond
This is an automated email from the ASF dual-hosted git repository.
wuwei pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from 4d4346d [Doc] TVM_REGISTER_API -> TVM_REGISTER_GLOBAL (#4768)
add 1ae44cf Fix Tensorflow conv3d pad bug,
masahi opened a new pull request #4773: [QNN] Add missing nullptr check
URL: https://github.com/apache/incubator-tvm/pull/4773
I get segfault during development and I had to use gdb to know that segfault
came from nullptr access in requantize.
@anijain2305 @vinx13 please review
masahi commented on a change in pull request #4771: [Relay] Added Merge
Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771#discussion_r370447222
##
File path: python/tvm/relay/transform.py
##
@@ -508,6 +508,41 @@ def Legalize(legalize_map_attr_name="FTVM
tqchen merged pull request #4768: [Doc] TVM_REGISTER_API -> TVM_REGISTER_GLOBAL
URL: https://github.com/apache/incubator-tvm/pull/4768
This is an automated message from the Apache Git Service.
To respond to the message, pleas
tqchen commented on issue #4768: [Doc] TVM_REGISTER_API -> TVM_REGISTER_GLOBAL
URL: https://github.com/apache/incubator-tvm/pull/4768#issuecomment-577945839
Thanks @hlu!
This is an automated message from the Apache Git Service
This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from b9328d0 [VTA] Support network which have no unique operator as
start/stop name for graph pack. (#4703)
tqchen commented on a change in pull request #4765: [tir] make IterVar extends
Var
URL: https://github.com/apache/incubator-tvm/pull/4765#discussion_r370429075
##
File path: include/tvm/tir/expr.h
##
@@ -236,21 +228,33 @@ enum IterVarType : int {
* \brief Iteration Varia
tqchen commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577944754
I am half joking but who knows, maybe we will get there
This is an autom
alexgl-github commented on issue #4772: Fix Tensorflow conv3d pad bug, add
non-cubic data and kernel tests
URL: https://github.com/apache/incubator-tvm/pull/4772#issuecomment-577937448
@optima2005 Would you mind reviewing?
Th
masahi edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577930418
so tvm will finally compile itself :) didn't know self hosting is one of our
goal
--
masahi commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577930418
so tvm will finally compile itself :)
This is an automated message from
tqchen commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577929141
btw, we all love python. Perhaps the final solution is we use tvm to compile
these python descriptions XD so we don’t need to to have the
icemelon9 commented on issue #4644: [Draft] Relay op strategy
URL: https://github.com/apache/incubator-tvm/pull/4644#issuecomment-577928887
x86 conv2d is already in the pr. working on cuda and more targets now
This is an autom
MarisaKirisame commented on issue #4644: [Draft] Relay op strategy
URL: https://github.com/apache/incubator-tvm/pull/4644#issuecomment-577928609
When is conv2d gonna be here? Need it for training.
This is an automated message
masahi commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577927317
will wait for others to give opinions, will merge tomorrow.
This is an
comaniac commented on issue #4465: [AutoTVM] Tune softmax CUDA schedule
URL: https://github.com/apache/incubator-tvm/pull/4465#issuecomment-577923200
@Laurawly maybe we could help to test it on a p3 instance so that we can get
this merged?
--
tqchen edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577921569
We can have another discussion thread about the scope of ops that goes to
cxx.
My current take is that the decision will l
tqchen commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577921569
We can have another discussion thread about the scope of ops that goes to
cxx.
My current take is that the decision will likely be
soiferj closed pull request #4465: [AutoTVM] Tune softmax CUDA schedule
URL: https://github.com/apache/incubator-tvm/pull/4465
This is an automated message from the Apache Git Service.
To respond to the message, please log on
soiferj commented on issue #4465: [AutoTVM] Tune softmax CUDA schedule
URL: https://github.com/apache/incubator-tvm/pull/4465#issuecomment-577917547
Hi all, I am not sure I'll be able to complete this PR at the moment. It is
hard for me to test, since I only have a Windows machine with GPU,
icemelon9 commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577910874
I agree with @masahi. During the work of #4644, I found a few inconsistency
between cpp and python topi implementation. Even for some
alexgl-github opened a new pull request #4772: Fix Tensorflow conv3d pad bug,
add non-cubic data and kernel tests
URL: https://github.com/apache/incubator-tvm/pull/4772
Thanks for contributing to TVM! Please refer to guideline
https://docs.tvm.ai/contribute/ for useful information and ti
masahi commented on issue #4756: [Docker] Update torch version to 1.4
URL: https://github.com/apache/incubator-tvm/pull/4756#issuecomment-577900368
hmm no idea why CI is failing. @leandron do you know what is going with
sphinx issues?
---
tmoreau89 commented on issue #4703: [VTA] Support network which have no unique
operator as start/stop name for graph pack.
URL: https://github.com/apache/incubator-tvm/pull/4703#issuecomment-577899433
Thank you @huajsj @liangfu , the PR has been merged.
tmoreau89 merged pull request #4703: [VTA] Support network which have no unique
operator as start/stop name for graph pack.
URL: https://github.com/apache/incubator-tvm/pull/4703
This is an automated message from the Apache
This is an automated email from the ASF dual-hosted git repository.
moreau pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from 23ba37d pooling.cc improvements (#4767)
add b9328d0 [VTA] Support network which have no unique operator
masahi edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577882904
Choosing cpp vs python per operator (rather than one lang for all ops) is a
good compromise, but it creates other problems (need
masahi edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577882904
Choosing cpp vs python per operator (rather than one lang for all ops) is a
good compromise, but it creates other problems (need
masahi edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577882904
Choosing cpp vs python per operator (rather than one lang for all ops) is a
good compromise, but it creates other problems (need
masahi edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577882904
Choosing cpp vs python per operator (rather than one lang for all ops) is a
good compromise, but it creates other problems (need
masahi commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577882904
Choosing cpp vs python per operator (rather than one lang for all ops) is a
good compromise, but it creates other problems (need to discu
tqchen edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577874461
Hmm, i hear you. I think the nature of “need to adding new features” marks
the op unstable, thus merit a python version as the pr
tqchen edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577874461
Hmm, i hear you. I think the nature of “need to adding new features” marks
the op unstable, thus merit a python version as the pr
tqchen commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577874461
Hmm, i hear you. I think the nature of “need to adding new features” marks
the op unstable, thus merit a python version as the primary on
masahi edited a comment on issue #4769: [TOPI] Remove cpp upsampling and resize
op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577870143
I do agree that the cpp solution is desirable and has many benefits. For
operators whose api and implementation would not be expe
masahi commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577870143
I do agree that the cpp solution is desirable and has many benefits. For
operators whose api and implementation would not be expected to
huajsj commented on a change in pull request #4703: [VTA] Support network which
have no unique operator as start/stop name for graph pack.
URL: https://github.com/apache/incubator-tvm/pull/4703#discussion_r370328795
##
File path: vta/python/vta/top/graphpack.py
##
@@ -309,
tmoreau89 commented on a change in pull request #4703: [VTA] Support network
which have no unique operator as start/stop name for graph pack.
URL: https://github.com/apache/incubator-tvm/pull/4703#discussion_r370317355
##
File path: vta/python/vta/top/graphpack.py
##
@@ -3
tqchen commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577710438
I do not mean that we have do it in this PR. Indeed having a C++ impl would
be harder than the python version.
On the other hand,
masahi commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577656902
I don't think C++ only solution is realistic so I prefer to keep only
python. C++ resize in particular is hopelessly outdated and I don't
mbarrett97 opened a new pull request #4771: [Relay] Added Merge Composite pass
URL: https://github.com/apache/incubator-tvm/pull/4771
This adds a pass, MergeComposite, which merges patterns (expressed as Relay
expressions). The detected patterns are wrapped in a function call which is
mark
tqchen commented on issue #4769: [TOPI] Remove cpp upsampling and resize op
URL: https://github.com/apache/incubator-tvm/pull/4769#issuecomment-577646627
In the case of upsampling and resize though should we consider them as
stable ops and keep only cxx impl instead ?
I agree it is b
masahi commented on a change in pull request #4497: [WIP] [Relay] Add a PyTorch
to Relay Parser
URL: https://github.com/apache/incubator-tvm/pull/4497#discussion_r370061548
##
File path: python/tvm/relay/frontend/pytorch.py
##
@@ -0,0 +1,1138 @@
+# Licensed to the Apache S
pingsutw opened a new pull request #4770: Bump prebuilt-image version in demo
dockerfile
URL: https://github.com/apache/incubator-tvm/pull/4770
Update prebuild image version to v0.55 due to v0.54 can't find
ppa:jonathonf/python-3.6
This ppa has been removed from public.
https://launchp
masahi commented on a change in pull request #4497: [WIP] [Relay] Add a PyTorch
to Relay Parser
URL: https://github.com/apache/incubator-tvm/pull/4497#discussion_r370061548
##
File path: python/tvm/relay/frontend/pytorch.py
##
@@ -0,0 +1,1138 @@
+# Licensed to the Apache S
masahi commented on a change in pull request #4497: [WIP] [Relay] Add a PyTorch
to Relay Parser
URL: https://github.com/apache/incubator-tvm/pull/4497#discussion_r370042351
##
File path: python/tvm/relay/frontend/pytorch.py
##
@@ -0,0 +1,1138 @@
+# Licensed to the Apache S
masahi commented on issue #4756: [Docker] Update torch version to 1.4
URL: https://github.com/apache/incubator-tvm/pull/4756#issuecomment-577617857
ok reverted
This is an automated message from the Apache Git Service.
To respo
leandron commented on issue #4756: [Docker] Update torch version to 1.4
URL: https://github.com/apache/incubator-tvm/pull/4756#issuecomment-577603634
Yes @masahi. Then, if you want, you can remove those `<7` in Pillow
installations, in the context of this PR, so it is a self-contained torch
masahi commented on issue #4497: [WIP] [Relay] Add a PyTorch to Relay Parser
URL: https://github.com/apache/incubator-tvm/pull/4497#issuecomment-577579947
@alexwong I'm working on loading quantized pytorch models to Relay via QNN.
Your work gave me a great head start :)
--
alexwong commented on issue #4497: [WIP] [Relay] Add a PyTorch to Relay Parser
URL: https://github.com/apache/incubator-tvm/pull/4497#issuecomment-577564701
@masahi Thanks for the help on this! I will address the comments and resume
work on this tomorrow.
-
68 matches
Mail list logo