Trainer is always in waiting state

Sentdex / Carla-RL

Reinforcement Learning codebase for self-driving car in Carla

MIT License

364 stars 98 forks source link

Trainer is always in waiting state #3

Open PankajJain5889 opened 5 years ago

PankajJain5889 commented 5 years ago

I just started using this after initial setup done right I am able to see agent output and NPc's but the trainer is always in waiting state

harbarex commented 4 years ago

@PankajJain5889 did you find a solution for this?

PankajJain5889 commented 4 years ago

Couldn't find a viable solution but will look in it this weekend, also what version of python and tensorflow-gpu are you using

harbarex commented 4 years ago

Try installing the latest version of cudnn. And change the value of TRAINER_MEMORY_ALLOCATION to 0.4. I am using the following configuration, and the code worked for me:

tensorflow-gpu: 1.15.0
keras: 2.2.4
cuda: 10.0.1
cudnn: 7.6.4
Python: 3.7.5

PankajJain5889 commented 4 years ago

Can you tell me what graphic card are you working with I am just unsure that 10.0.1 Cuda would work with mine cause as per my online research 9.0 is stable, please suggest

harbarex commented 4 years ago

I am using the NVIDIA Quadro P4000 - 8GB Memory

FHof commented 4 years ago

I've noticed that the ConsoleStats hides a python keras Exception and found a way to avoid it: #9