#!/bin/bash #export TPU_IP_ADDRESS=10.8.195.90; #export XRT_TPU_CONFIG="tpu_worker;0;$TPU_IP_ADDRESS:8470" #export TPU_NAME="grpc://$TPU_IP_ADDRESS:8470" export XRT_WORKERS="localservice:0;grpc://localhost:40934" export XRT_DEVICE_MAP="CPU:0;/job:localservice/replica:0/task:0/device:XLA_CPU:0|GPU:0;/job:localservice/replica:0/task:0/device:XLA_GPU:0" #export PYTHONPATH=$SCRATCH/:${PYTHONPATH} #export PYTHONPATH=/gpfsscratch/rech/imi/usc19dv/lib/python3.7/site-packages:${PYTHONPATH} module load pytorch-gpu/py3/1.8.0 py=python3 #root_dir=$SCRATCH/data root_dir=data ####aistpp_60hz #data_dir=${root_dir}/scaled_features #hparams_file=aistpp_60hz/transflower_aistpp_expmap #hparams_file=aistpp_60hz/transglower_aistpp_expmap ####aistpp_20hz #data_dir=${root_dir}/aistpp_20hz #exp=$1 #exp=transglower_aistpp_expmap #exp=transglower_residual_aistpp_expmap #exp=transflower_residual_aistpp_expmap #exp=mowgli_aistpp_expmap #exp=transflower_aistpp_expmap #exp=transflower_residual_aistpp_expmap #exp=residualflower2_transflower_aistpp_expmap #exp=moglow_aistpp_expmap #hparams_file=aistpp_20hz/${exp} #exp=mowgli_aistpp_expmap_future3 ## Fix: needs vmapped version of transformer: #hparams_file=aistpp_20hz/residualflower2_moglow_aistpp_expmap ####moglow_pos #data_dir=${root_dir}/moglow_pos #exp=$1 #exp=transglower_moglow_pos #exp=transglower_residual_moglow_pos #exp=transflower_residual_moglow_pos #exp=transflower_moglow_pos #exp=residualflower2_transflower_moglow_pos #exp=moglow_moglow_pos #exp=moglow_trans_moglow_pos #hparams_file=moglow_pos/${exp} #exp=testing #exp=${exp}_pos_emb ####dance_combined #data_dir=${root_dir}/dance_combined data_dir=${root_dir}/dance_combined2 #data_dir=${root_dir}/dance_combined3 exp=$1 #exp=transformer_expmap #exp=mowgli_expmap_stage2 #exp=mowgli_expmap #exp=transflower_expmap_cr4 #exp=transglower_aistpp_expmap #exp=transglower_residual_aistpp_expmap #exp=transflower_residual_aistpp_expmap #exp=transflower_aistpp_expmap #exp=residualflower2_transflower_aistpp_expmap #exp=moglow_aistpp_expmap hparams_file=dance_combined/${exp} #exp=${exp}_future3_actnorm #exp=${exp}_future3 exp=testing #exp=mowgli_expmap_stage2_newdata2 #exp=testing2 echo $exp $py training/train.py --data_dir=${data_dir} --max_epochs=2000\ --hparams_file=training/hparams/${hparams_file}.yaml \ --experiment_name=$exp\ --accelerator=ddp \ --workers=0 \ --gpus=1 \ ${@:2} #NOTE: can override experiment_name, and any of the options above #--batch_size=8 \ #--continue_train \ #--no_load_hparams \ #--load_weights_only \ #--only_load_in_state_dict=vae \ #--fix_lengths \ #--continue_train \ #--workers=$(nproc) \ #--gpus=1 \ #--scales="[[16,0]]" \ #--flow_dist=studentT \ #--output_lengths="3" \ # --stage2 \ # --continue_train \ # --load_weights_only \ # --prior_use_x_transformers \ # --use_x_transformers \ # --use_rotary_pos_emb \ # --learning_rate=1e-5 \ #--residual_scales="[[16,0]]" # --glow_norm_layer="actnorm" \ #--use_pos_emb_output \ # --tpu_cores=8 \