landersson commented on issue #8388: Exit race condition on library handle usage
URL:
https://github.com/apache/incubator-mxnet/issues/8388#issuecomment-338879101
I guess calling exit(0) before reaching the end changes the order in which
objects are deleted. I can't reproduce the error
cjolivier01 commented on issue #8388: Exit race condition on library handle
usage
URL:
https://github.com/apache/incubator-mxnet/issues/8388#issuecomment-338876038
This sort of error doesn?t look familiar. What?s the difference in behavior
of just letting the script finish rather than
qiliux opened a new issue #8404: gluon.loss.SoftmaxCrossEntropyLoss() missed
ignore_label parameter?
URL: https://github.com/apache/incubator-mxnet/issues/8404
I found the `mx.symbol. SoftmaxOutput` has parameter ignore_label, however
i can't find `gluon.loss.SoftmaxCrossEntropyLoss` has
eric-haibin-lin opened a new pull request #8403: [WIP] dtype default to
source_array.dtype for sparse ndarrays
URL: https://github.com/apache/incubator-mxnet/pull/8403
## Description ##
csr_matrix(source) always creates a new CSRNDArray with dtype=float32
instead of using source.dtype,
qingzhouzhen commented on issue #8389: dpn feature request
URL:
https://github.com/apache/incubator-mxnet/issues/8389#issuecomment-338850534
OK, give me some time @chinakook
This is an automated message from the Apache Git
junranhe closed issue #8290: dist_sync kvstore can not load balance ?
URL: https://github.com/apache/incubator-mxnet/issues/8290
This is an automated message from the Apache Git Service.
To respond to the message, please log
qiliux closed issue #8396: reimplement fcn-xs example using gluon
URL: https://github.com/apache/incubator-mxnet/issues/8396
This is an automated message from the Apache Git Service.
To respond to the message, please log on
qiliux commented on issue #8396: reimplement fcn-xs example using gluon
URL:
https://github.com/apache/incubator-mxnet/issues/8396#issuecomment-338841360
@piiswrong thank you. I will try it. If i succeed, i will update the answer.
larroy opened a new pull request #8402: Lint fix
URL: https://github.com/apache/incubator-mxnet/pull/8402
## Description ##
Minor Lint fix
## Checklist ##
### Essentials ###
- [x] Passed code style checking (`make lint`)
- [x] Changes are complete (i.e. I finished coding
cjolivier01 commented on issue #8395: [CMAKE][ARM] USE_SSE option to govern
USE_SSE in mshadow and -msse2 compil?
URL: https://github.com/apache/incubator-mxnet/pull/8395#issuecomment-338839266
I just think it doesn't make logical sense. If @piiswrong likes it, then I
am fine, but why t
larroy opened a new pull request #8401: Make make lint compatible with python3
(don't call python2 explicitly)
URL: https://github.com/apache/incubator-mxnet/pull/8401
## Description ##
(Brief description on what this PR is about)
## Checklist ##
### Essentials ###
- [x]
larroy opened a new pull request #8400: Fixes #8292. Bug on Slice accessing
uninitialized memory in param.beg?
URL: https://github.com/apache/incubator-mxnet/pull/8400
?in. (Out of bounds array access).
## Checklist ##
### Essentials ###
- [x] Passed code style checking (`make
larroy commented on issue #8395: [CMAKE][ARM] USE_SSE option to govern USE_SSE
in mshadow and -msse2 compil?
URL: https://github.com/apache/incubator-mxnet/pull/8395#issuecomment-338827849
Can you be more specific in what's your alternative proposal?
larroy commented on issue #8395: [CMAKE][ARM] USE_SSE option to govern USE_SSE
in mshadow and -msse2 compil?
URL: https://github.com/apache/incubator-mxnet/pull/8395#issuecomment-338827626
I added the switch for consistency. I think it would be better to change
this sse to -march=native
larroy commented on issue #8395: [CMAKE][ARM] USE_SSE option to govern USE_SSE
in mshadow and -msse2 compil?
URL: https://github.com/apache/incubator-mxnet/pull/8395#issuecomment-338827386
Wheres the visual C++ flag set?
cjolivier01 commented on issue #8340: Optimizations to set_to, copy, fill ops
URL: https://github.com/apache/incubator-mxnet/pull/8340#issuecomment-338826252
Any further comments on this, or everyone ok with me merging once it passes
CI?
mseeger commented on issue #8361: Simplified unary/binary math operators
URL: https://github.com/apache/incubator-mxnet/pull/8361#issuecomment-338802232
@piiswrong Jenkins just won't run for this one, I am giving up
This is
ptrendx opened a new pull request #8399: Fix for "name epochs is not defined"
URL: https://github.com/apache/incubator-mxnet/pull/8399
## Description ##
Fix for "name epochs is not defined" when testing image_classification.py
with symbolic mode.
## Checklist ##
### Essentials
piiswrong commented on a change in pull request #8398: fix topk with big shape
URL: https://github.com/apache/incubator-mxnet/pull/8398#discussion_r146391780
##
File path: tests/python/unittest/test_ndarray.py
##
@@ -593,6 +593,10 @@ def gt_topk(dat, axis, ret_typ, k,
reminisce commented on issue #8366: MXNet training slowing down after several
epochs (inception_bn with 1 million images)
URL:
https://github.com/apache/incubator-mxnet/issues/8366#issuecomment-338785050
Does the symptom exist in using fewer GPUs? What type of kvstore did you use?
ZiyueHuang opened a new pull request #8398: fix topk with big shape
URL: https://github.com/apache/incubator-mxnet/pull/8398
## Description ##
fix https://github.com/apache/incubator-mxnet/issues/8303
@eric-haibin-lin
## Checklist ##
### Essentials ###
- [ ] Passed
moakra opened a new issue #8397: A3C with multiple workers
URL: https://github.com/apache/incubator-mxnet/issues/8397
Hi,
When I run the a3c from this
[repo](https://github.com/apache/incubator-mxnet/tree/master/example/reinforcement-learning/a3c)
using the file launcher.py, it can
piiswrong commented on issue #8371: Add note in the doc for using naive engine
in multithreading environment
URL: https://github.com/apache/incubator-mxnet/pull/8371#issuecomment-338745901
I think it's enough to simply say:
ThreadedEngine will deadlock after a fork because it is
piiswrong commented on issue #8396: reimplement fcn-xs example using gluon
URL:
https://github.com/apache/incubator-mxnet/issues/8396#issuecomment-338741357
You can implement the whole thing in a custom block, without using sequential
larroy commented on issue #8292: mx.nd.array indexing broken in armv7 /
raspberrypi / jessie 8.0 (5 dimensional tensor)
URL:
https://github.com/apache/incubator-mxnet/issues/8292#issuecomment-338707413
Found the bug:
in matrix_op-inl.h:472 we are addressing the parameter vector
ZiyueHuang commented on issue #8349: [New feature Request] bincount function
URL:
https://github.com/apache/incubator-mxnet/issues/8349#issuecomment-338707218
It's hard to implement the same function as `numpy.bincount` because the
shape of output depends on the data of input. Do you have
cjolivier01 commented on issue #8395: [CMAKE][ARM] USE_SSE option to govern
USE_SSE in mshadow and -msse2 compil?
URL: https://github.com/apache/incubator-mxnet/pull/8395#issuecomment-338706170
Doesn?t the SUPPORT_SSE test fail for platform which doesn?t support it?
cjolivier01 commented on issue #8395: [CMAKE][ARM] USE_SSE option to govern
USE_SSE in mshadow and -msse2 compil?
URL: https://github.com/apache/incubator-mxnet/pull/8395#issuecomment-338705239
Just curious what the use-case is for not wanting sse
qiliux commented on issue #8394: How to include Negative samples/images for
training
URL:
https://github.com/apache/incubator-mxnet/issues/8394#issuecomment-338689084
You can use some other type images to a third class.
Then use `im2rec` tool to make the dataset.
ZiyueHuang commented on a change in pull request #8331: [sparse] slice_axis for
csr, cpu implementation
URL: https://github.com/apache/incubator-mxnet/pull/8331#discussion_r146285154
##
File path: src/operator/tensor/matrix_op-inl.h
##
@@ -927,6 +927,140 @@ void
ZiyueHuang commented on a change in pull request #8331: [sparse] slice_axis for
csr, cpu implementation
URL: https://github.com/apache/incubator-mxnet/pull/8331#discussion_r146284549
##
File path: tests/python/unittest/test_sparse_ndarray.py
##
@@ -127,9 +127,27 @@ def
Yochengliu commented on issue #8385: Save weights at any time between integral
epochs
URL:
https://github.com/apache/incubator-mxnet/issues/8385#issuecomment-338524955
@solin319 Thank you, I will have a try, and do you know some specific
saving ways that just press 'Ctrl + C', then the
KellenSunderland commented on issue #8387: test_operator_gpu.test_cuda_rtc
failure - Jetson TX2 (GPU/CUDA)
URL:
https://github.com/apache/incubator-mxnet/issues/8387#issuecomment-338666572
@jinfagang Totally agree jinfagang, more documentation will be coming for
ARM devices. There's a
KellenSunderland commented on issue #8387: test_operator_gpu.test_cuda_rtc
failure - Jetson TX2 (GPU/CUDA)
URL:
https://github.com/apache/incubator-mxnet/issues/8387#issuecomment-338665104
@piiswrong The problem is that we're passing a default compute capability of
30 which works on p2
KellenSunderland opened a new issue #8387: test_operator_gpu.test_cuda_rtc
failure - Jetson TX2 (GPU/CUDA)
URL: https://github.com/apache/incubator-mxnet/issues/8387
## Description
The test test_operator_gpu.test_cuda_rtc is consistently failing on a Jetson
TX2.
## Environment
qiliux opened a new issue #8396: reimplement fcn-xs example using gluon
URL: https://github.com/apache/incubator-mxnet/issues/8396
Package used (Python/R/Scala/Julia):
(I'm using ...)
Python
My problem is that how to use the pre-layer's info in my custom layer to
implement
qiliux closed issue #8348: mxnet.gluon.data.vision.ImageRecordDataset key error
URL: https://github.com/apache/incubator-mxnet/issues/8348
This is an automated message from the Apache Git Service.
To respond to the message,
qiliux closed issue #8386: Does gluon have a crop layer (like caffe crop layer)
?
URL: https://github.com/apache/incubator-mxnet/issues/8386
This is an automated message from the Apache Git Service.
To respond to the
qiliux commented on issue #8386: Does gluon have a crop layer (like caffe crop
layer) ?
URL:
https://github.com/apache/incubator-mxnet/issues/8386#issuecomment-338657191
Summary design your own gluon layer:
```
class CenteredLayer(Block):
def __init__(self, **kwargs):
jinfagang commented on issue #8387: test_operator_gpu.test_cuda_rtc failure -
Jetson TX2 (GPU/CUDA)
URL:
https://github.com/apache/incubator-mxnet/issues/8387#issuecomment-338655061
Is there any tutorial or blog indicates how to build mxnet on tegra TX2, I
really hope mxnet can be built
larroy opened a new pull request #8395: [CMAKE] USE_SSE option to govern
USE_SSE in mshadow and -msse2 compil?
URL: https://github.com/apache/incubator-mxnet/pull/8395
?er flags
Updated mshadow.
## Description ##
This fixes Cmake build on ARM. (NO SSE support nor X86
SumNeuron opened a new issue #8393: nd.concat with empty array throws error
URL: https://github.com/apache/incubator-mxnet/issues/8393
So I want to initialize an empty array and throw some data in it over a
training loop.
However, doing so causes a very odd error:
dumb =
tornadomeet opened a new issue #8392: symbol/operator type feature request
URL: https://github.com/apache/incubator-mxnet/issues/8392
Hello, currently we cannot get the type of one operator, such as we need a
function named `get_type()`
```
a = mx.sym.Convolution(?)
b =
HaoLiuHust opened a new issue #8391: use 224x224 size to train mobilenet will
encounter memory problem
URL: https://github.com/apache/incubator-mxnet/issues/8391
I was trying to use CIFAR dataset to train a mobilenet, since the origin
size is 32x32 so I resize them to 224x224 which is the
rahul003 commented on a change in pull request #8342: [WIP] 2bit gradient
compression
URL: https://github.com/apache/incubator-mxnet/pull/8342#discussion_r146181250
##
File path: src/operator/contrib/two_bit_quantize-inl.h
##
@@ -0,0 +1,340 @@
+/*
+ * Licensed to the
rahul003 commented on a change in pull request #8342: [WIP] 2bit gradient
compression
URL: https://github.com/apache/incubator-mxnet/pull/8342#discussion_r146181250
##
File path: src/operator/contrib/two_bit_quantize-inl.h
##
@@ -0,0 +1,340 @@
+/*
+ * Licensed to the
blakeliu commented on issue #8339: data iterators tutorial errors
URL:
https://github.com/apache/incubator-mxnet/issues/8339#issuecomment-338567083
same error
ValueErrorTraceback (most recent call last)
in ()
9
10 mod =
jiarenyf commented on issue #8347: CTC Example Problem
URL:
https://github.com/apache/incubator-mxnet/issues/8347#issuecomment-338561205
??
This is an automated message from the Apache Git Service.
To respond to the message,
zachahuy commented on issue #8390: ValueError: need more than 1 value to unpack
URL:
https://github.com/apache/incubator-mxnet/issues/8390#issuecomment-338558572
I also tried this and got this error.
mod = mx.mod.Module(
49 matches
Mail list logo
