I have a least one week in try to use my CUDA to train my rasa project. But, today I looked that the rasa --version command displayed the next message:
C:\blife\ai\chatbot\venv\lib\site-packages\rasa\shared\utils\io.py:99: UserWarning: You have an environment variable 'TF_GPU_MEMORY_ALLOC' set but no GPUs were detected to configure.
Rasa Version : 3.6.17
Minimum Compatible Version: 3.5.0
Rasa SDK Version : 3.6.2
Python Version : 3.10.4
Operating System : Windows-10-10.0.22631-SP0
Python Path : C:\blife\ai\chatbot\venv\Scripts\python.exe
Where says there is not GPU in my machine.
What are the steps to use GPU and CUDA to train and run RASA?
Also to install CUDA toolkit but it launches an error where NVIDIA Nsight for Visual Studio 2017 is required to complete the install process. I cannot find Visual Studio 2017 Community Edition. Only I found 2019 ans 2022 version.
To use your computer’s graphics card (GPU) to make Rasa run faster, follow these steps: First, make sure your computer recognizes your GPU. If you get an error saying it doesn’t, check that your GPU drivers are installed correctly. Next, adjust a setting called ‘TF_GPU_MEMORY_ALLOC’ to help TensorFlow (the software Rasa uses) use your GPU properly. You can find how to do this in the TensorFlow documentation. Then, make sure you have the latest drivers for your GPU from Nvidia’s website. If you have trouble, check if your computer’s other drivers need updating too. After that, install something called the CUDA Toolkit, which helps with GPU acceleration for Rasa.
If you run into problems, make sure the CUDA Toolkit is compatible with your computer and GPU, and look at the CUDA Toolkit documentation for help. Lastly, if you need to install a tool called Nsight for Visual Studio 2017 for the CUDA Toolkit, consider trying newer versions like 2019 or 2022. You can also check Nvidia’s developer website for other tools that might work better.