-
Notifications
You must be signed in to change notification settings - Fork 35
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Missing Modules for C_HLS Export #5
Comments
Hello, While N2D2 is our deep learning open-source core framework, some modules referred as "N2D2-IP" in the manual, are only available through custom license agreement with CEA LIST. If you are interested in obtaining some of these modules, please contact our business developer for more information on available licensing options: Sandrine VARENNE (Sandrine.VARENNE@cea.fr) In addition to N2D2-IP modules, we can also provide our expertise to design specific solutions for integrating DNN in embedded hardware systems, where power, latency, form factor and/or cost are constrained. We can target CPU/DSP/GPU CoTS hardware as well as our own PNeuro (programmable) and DNeuro (dataflow) dedicated hardware accelerator IPs for DNN on FPGA or ASIC. This is now clarified in the new "About N2D2-IP" section in the manual (rev 8a294d7). |
Changing init -> make_compositable fixed the issue.
## Context The Python API initializes several times deepNets with the Keras Interop. However, it is an impossible behaviour with the MultiGPU. The objective of this MR is to fix this issue. ## Issue Some logs about the issue: ``` CUBLAS initialized on device #5 Cuda failure: peer access is already enabled (704) in n2d2/src/DeepNet.cpp:1027 on device #7 ### Training ### # Train Epoch: 0 # Traceback (most recent call last): File "../quantizer_ONNX.py", line 153, in <module> x = provider.read_random_batch() File "n2d2/python/n2d2/provider.py", line 186, in read_random_batch self._deepnet.N2D2().initialize() RuntimeError: Cuda failure: peer access is already enabled (704) in n2d2/src/DeepNet.cpp:1027 on device #7 ``` To speed up the communications among devices, there is a CUDA option to enable peer accesses among devices when it is possible. The activation of this option happens during the initialization of the deepNet but should only happen once. With the Python API, the initialization is executed more than one time and creates this issue. ## Solution To fix the problem, the "enable function" has been moved to CudaUtils and should be executed just at the beginning of the program. Moreover, if the function is called more than one time, a warning will be raised.
Hi. I'm trying to use the tool to export the trained model into C_HLS codes as shown in the manual. However the tool prompts me that additional modules are needed for the export. I wonder if the module for C_HLS export is missing? It seems the only available export right now is CPP_cuDNN. Thanks.
The text was updated successfully, but these errors were encountered: