dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-638231002
https://gist.github.com/dcslin/8fe77d3a33b0a7406012daa6304084c1
CUDNN LSTM API current and new, working on updating the corresponding code.
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-637918244
> @dcslin Hi, shicong, for some cases of Conv2d, there may be some issues,
please use this code to test:
>
> ```
> def _conv2d_helper(self, dev):
> #
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-636821604
As discussed, last commit changes: 1 Restore creator settings for interim
tensors; 2 update device_check() after new get states implementation; 3 fix
typo in LSTM; 4 fix test case
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-635127163
Hi @joddiy , could you help to test onnx, it it working at my side now.
This is an automated message from the Apache
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-633771962
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-633499903
> > there is no params explicitly defined by MyModel; therefore there should
be no direct params of MyModel. All params are from the sublayers of MyModel,
i.e., l1 and l2.
>
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-633374596
> there is no params explicitly defined by MyModel; therefore there should
be no direct params of MyModel. All params are from the sublayers of MyModel,
i.e., l1 and l2.
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-633259477
last commit: allow list of tensor as input of initialize()
Fix conv2d in channels check
Fix rnn_base override method
Fix LSTM rnn initialise args, __init__ method, params
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-633231970
One of the caveat worth noting for user is that args of `initialize()`,
`forward()` need to match exactly to `__call__()`
Normally `__call__(x)` is the same as `initialize(x)`,
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-632984025
setup the overall framework for save load states
This is an automated message from the Apache Git Service.
To
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-632811131
consolidating some designs of layer method with rulin
TODO:
5. [rulin & shicong] compile / do_init: separate the initialzation and
forward propagation for each layers, create
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-632461085
sorry did not push the commit timely as I tried to minimise the commits
number
This is an automated message from
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-632460967
I will compare my version and rulin and see how..
https://github.com/dcslin/singa/tree/py-api
https://github.com/dcslin/singa/tree/py-api2
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-632460336
> > > TODO list compiled by @XJDKC with slight update:
> > >
> > > * [x][rulin] 1. separate layers from autograd.py to form a new
file(`layer.py`)
> > > _Alias added
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-632459592
TODO list compiled by @XJDKC with slight update:
* [x] [rulin] 1. separate layers from autograd.py to form a new
file(`layer.py`)
_Alias added in `autograd.py` for
dcslin commented on pull request #697:
URL: https://github.com/apache/singa/pull/697#issuecomment-630741700
Hi @nudles @nudles get/set params are implemented as required:
```
{'l1:W': [[-0.8381560.58935565]
[ 0.4453408 0.08828315]
[ 0.10353903 0.974195 ]
[
16 matches
Mail list logo