TFAMI.v4 Upcoming Release
ritchieng opened this issue · comments
Name Change:
- TFAMI would be changed to DLAMI (dee-luh-mi) to include PyTorch and TensorFlow (deep learning frameworks).
Essential features:
- TensorFlow r1.0
- Latest PyTorch
- Latest Keras
- Latest TensorLayer
- CUDA 8.0
- CuDNN 5.1
- Python 2.7
- Ubuntu 16.04
Community requested changes
@ritchieng can we add http://pytorch.org/
It is simple to install
@dev-rkoshlyak Yeah, I've been wanting to add that. I would probably have to change the name from TFAMI to DLAMI to include that.
You can currently just install it yourself with a simple pip command as the environment is made for it :)
any possibility of python 3.5?
@mjdietzx I'll actually consider that. Let me think :)
Thanks for your support everyone.
However I found out that our funding from Amazon expired and these AMIs cost us money across the regions. As such we would be discontinuing these AMIs until we've funding from Amazon. :(
Ritchie,
Maybe you can recruit other Githubers all over the Amazon regions to host your AMI. For instance, I am in EU-Ireland, so I could host it... If I remember well, the cost is just the one corresponding to storage, isn't it?
That will be a pity for me if I no longer have access to your AMI since it saves my time not to install, I have some difficulties using TF on GPU for unknown reasons
@mphuget +1
@Innarticles @mphuget I'm discussing with my other researcher to host on in one location :) It's really a heartbreak to close this too.
In the mean time I'll be hosting it in one location.
@ritchieng Do you think it is possible to reproduce the process you made to setup the AMI? I tried many, many sites proposing process to install CUDA, CuDNN, TF and for most of them, it is not working properly, especially on TF that evolves too quickly
@mphuget They're a lot of environment-specific debugging to be done that's why I can't release a standard one (tried, failed). That's the pain here. Even using a containerized installation, it's problematic. This is why I started TFAMI so everyone can just get to the deep learning part and skip the mess in installing.
@mphuget @Innarticles It's in only one region for now which has a lot of spot instances now. I sincerely apologize for the downtime for this AMI.