generate.py blows up GPU memory

[soumith]

When I ran language model example, generate.py blew up GPU memory as it generated sentences (starting from ~500MB to ~4GB). In the end I got out of memory error: RuntimeError: cuda runtime error (2) : out of memory at /data/users/soumith/miniconda2/conda-bld/pytorch-0.1.5_1479441063232/work/torch/lib/THC/generic/THCStorage.cu:65 .

Some info:

• Ubuntu 14.04
• Geforce GTX 970 (Memory size: 4GB)
• Python 3.5.2

cc: @adamlerer

[colesbury]

If you're doing inference only, you might need to set volatile=True. Otherwise you often end up with reference cycles which aren't collected immediately. (This is something we should fix in PyTorch)

[soumith]

fixed via 4c93238