search

facebookresearch / vip

Official repository for "VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Training"
https://sites.google.com/view/vip-rl
Other
135 stars 17 forks source link

Getting Segmentation fault (core dumped) at this line of trainer.py: alles = model(b_im_r) #7

Open mrshetab opened 10 months ago

mrshetab commented 10 months ago

I want to train vip on my own data, however, I get the segmentation fault error at this line of trainer.py: alles = model(b_im_r) The shape of b_im_r is torch.Size([128, 3, 224, 224]). here is my config file:

defaults:

snapshot

save_snapshot: false load_snap: ""

replay buffer

num_workers: 10 batch_size: 32 train_steps: 10001 eval_freq: 1000

misc

seed: 1 device: cuda

experiment

experiment: train_vip wandbuser: wandbproject:

data

dataset: dataset datapath: /home/reza/Lyon/Codes/action_recognition/vip/vip/dataset doaug: "none"

agent: target: vip.VIP device: ${device} lr: 1e-4 hidden_dim: 2 size: 34 l2weight: 0.001 l1weight: 0.001 num_negatives: 3

Any idea why this happens?

mrshetab commented 10 months ago

Problem solved, it was an issue of CUDA inconsistency. Thank you.