chatbot-rnn
chatbot-rnn copied to clipboard
Training on GPU?
What changes do I have to make to train.py to train using my GTX-1080 (I have tensorflow-gpu installed and recognizing it)?
You'll probably need to scale down the size of the net or the batch size or sequence length to accommodate that the GTX-1080 has only 8gb of VRAM instead of the 11gb of the 1080Ti that I trained on. I'd probably recommend scaling down the size of the net, for example python3 train.py --block_size 1024
.
I am trying to train the Chatbot on a , Core I5-3570, 12 gb RAM, GTX 1060 6gb any idea what I need to change to makes it run effectively? I have tried change the block_size but still somehow still training on my CPU , if I can train using both CPU + GPU what parameters should I add to it?
nvm i found a fix using Pip install --ignore-installed --upgrade tensorflow-gpu
FWIW: I'm training a bot instance with a 100MB input-set with a GeForce GTX 1050 Ti and this succeeds if I use --num_blocks 1 for a train.py parameter.