Closed ivancruzbht closed 7 years ago
Does this example require too much VRAM? I am trying to run this example on a box with a Nvida 970m (3GB VRAM) but I got this error:
05/16/2017 04:02:32 PM: [ Ok, let's go... ] 05/16/2017 04:02:32 PM: [ Training for 1000 iters... ] 05/16/2017 04:02:36 PM: [train] updates = 10 | train loss = 9.83 | exs = 310 05/16/2017 04:02:39 PM: [train] updates = 20 | train loss = 9.79 | exs = 623 05/16/2017 04:02:42 PM: [train] updates = 30 | train loss = 9.75 | exs = 938 THCudaCheck FAIL file=/py/conda-bld/pytorch_1493680494901/work/torch/lib/THC/generic/THCStorage.cu line=66 error=2 : out of memory Traceback (most recent call last): File "examples/drqa/train.py", line 178, in <module> main(opt) File "examples/drqa/train.py", line 113, in main train_world.parley() File "/home/ivan/ParlAI/parlai/core/worlds.py", line 505, in parley batch_act = self.batch_act(index, batch_observations[index]) File "/home/ivan/ParlAI/parlai/core/worlds.py", line 479, in batch_act batch_actions = a.batch_act(batch_observation) File "/home/ivan/ParlAI/parlai/agents/drqa/agents.py", line 192, in batch_act self.model.update(batch) File "/home/ivan/ParlAI/parlai/agents/drqa/model.py", line 113, in update self.optimizer.step() File "/home/ivan/anaconda3/lib/python3.6/site-packages/torch/optim/adamax.py", line 68, in step torch.max(norm_buf, 0, out=(exp_inf, exp_inf.new().long())) RuntimeError: cuda runtime error (2) : out of memory at /py/conda-bld/pytorch_1493680494901/work/torch/lib/THC/generic/THCStorage.cu:66
I haven't dive in the code, so I am not sure if this is a bug or I just need more VRAM. Thank you
I just changed the batch size to make it work.
Does this example require too much VRAM? I am trying to run this example on a box with a Nvida 970m (3GB VRAM) but I got this error:
I haven't dive in the code, so I am not sure if this is a bug or I just need more VRAM. Thank you