[GitHub] [incubator-mxnet] marcoabreu commented on issue #9686: [Discussion] MXNet 2.0 Roadmap (was: APIs that might be a good idea to break in 2.0)
marcoabreu commented on issue #9686: URL: https://github.com/apache/incubator-mxnet/issues/9686#issuecomment-664051201 Are there any plans to move the training logic (dataset handling, distributed training, etc) into the core to avoid having all that logic in the frontend languages? 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [incubator-mxnet] marcoabreu commented on issue #9686: [Discussion] MXNet 2.0 Roadmap (was: APIs that might be a good idea to break in 2.0)
marcoabreu commented on issue #9686: [Discussion] MXNet 2.0 Roadmap (was: APIs that might be a good idea to break in 2.0) URL: https://github.com/apache/incubator-mxnet/issues/9686#issuecomment-513004229 Could we get rid of all the different pre-processor statements in the codebase that evolved due to the different accelerators (USE_CUDA, USE_TVM, USE_MKLDNN, etc) and fully replace them with the accelerator API from @samskalicky ? This would heavily improve the maintainability. In terms of operator definitions, we could use ONNX as standard (or derive from it if it's not sufficient). At the moment, there's a tight coupling between the operator definitions and the accelerator choice. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] marcoabreu commented on issue #9686: [Discussion] MXNet 2.0 Roadmap (was: APIs that might be a good idea to break in 2.0)
marcoabreu commented on issue #9686: [Discussion] MXNet 2.0 Roadmap (was: APIs that might be a good idea to break in 2.0) URL: https://github.com/apache/incubator-mxnet/issues/9686#issuecomment-513002803 Agree, the entire API (at the C-API level) should be designed to be entirely threadsafe for all requests - whether it's inference or training. This includes parallel calls from different threads - speak no locking or sticky threads. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services