search

pender / chatbot-rnn

A toy chatbot powered by deep learning and trained on data from Reddit
MIT License
900 stars 370 forks source link

Training on GPU? #53

Open jesseos opened 5 years ago

jesseos commented 5 years ago

What changes do I have to make to train.py to train using my GTX-1080 (I have tensorflow-gpu installed and recognizing it)?

pender commented 5 years ago

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.

claudetheboof commented 5 years ago

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?

claudetheboof commented 5 years ago

nvm i found a fix using Pip install --ignore-installed --upgrade tensorflow-gpu

ghost commented 5 years ago

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.