add v1 hr subset of aesthetics training, resume v3

scripts/slurm/v1_iahr_torch111/launcher.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+
+# mpi version for node rank
+H=`hostname`
+THEID=`echo -e $HOSTNAMES  | python3 -c "import sys;[sys.stdout.write(str(i)) for i,line in enumerate(next(sys.stdin).split(' ')) if line.strip() == '$H'.strip()]"`
+export NODE_RANK=${THEID}
+echo THEID=$THEID
+
+echo "##########################################"
+echo MASTER_ADDR=${MASTER_ADDR}
+echo MASTER_PORT=${MASTER_PORT}
+echo NODE_RANK=${NODE_RANK}
+echo WORLD_SIZE=${WORLD_SIZE}
+echo "##########################################"
+# debug environment worked great so we stick with it
+# no magic there, just a miniconda python=3.9, pytorch=1.12, cudatoolkit=11.3
+# env with pip dependencies from stable diffusion's requirements.txt
+eval "$(/fsx/stable-diffusion/debug/miniconda3/bin/conda shell.bash hook)"
+#conda activate stable
+conda activate torch111
+cd /fsx/stable-diffusion/stable-diffusion
+
+CONFIG="/fsx/stable-diffusion/stable-diffusion/configs/stable-diffusion/v1_improvedaesthetics.yaml"
+
+# resume and set new seed to reshuffle data
+EXTRA="--seed 718 model.params.ckpt_path=/fsx/stable-diffusion/stable-diffusion/checkpoints2/v1pp/v1pp-flatline.ckpt"
+
+# only images >= 512
+EXTRA="${EXTRA} data.params.min_size=512"
+
+# time to decay
+#EXTRA="${EXTRA} model.params.scheduler_config.params.cycle_lengths=[50000] model.params.scheduler_config.params.f_min=[1e-6]"
+
+# custom logdir
+#EXTRA="${EXTRA} --logdir rlogs"
+
+# debugging
+#EXTRA="${EXTRA} -d True lightning.callbacks.image_logger.params.batch_frequency=50"
+
+/bin/bash /fsx/stable-diffusion/stable-diffusion/scripts/test_gpu.sh
+
+python main.py --base $CONFIG --gpus 0,1,2,3,4,5,6,7 -t --num_nodes ${WORLD_SIZE} --scale_lr False $EXTRA

scripts/slurm/v1_iahr_torch111/sbatch.sh

@@ -0,0 +1,43 @@
+#!/bin/bash
+#SBATCH --partition=compute-od-gpu
+#SBATCH --job-name=stable-diffusion-v1-iahr-torch111
+#SBATCH --nodes 20
+#SBATCH --ntasks-per-node 1
+#SBATCH --cpus-per-gpu=4
+#SBATCH --gres=gpu:8
+#SBATCH --exclusive
+#SBATCH --output=%x_%j.out
+#SBATCH --comment "Key=Monitoring,Value=ON"
+
+module load intelmpi
+source /opt/intel/mpi/latest/env/vars.sh
+export LD_LIBRARY_PATH=/opt/aws-ofi-nccl/lib:/opt/amazon/efa/lib64:/usr/local/cuda-11.0/efa/lib:/usr/local/cuda-11.0/lib:/usr/local/cuda-11.0/lib64:/usr/local/cuda-11.0:/opt/nccl/build/lib:/opt/aws-ofi-nccl-install/lib:/opt/aws-ofi-nccl/lib:$LD_LIBRARY_PATH
+#export LD_LIBRARY_PATH=/opt/aws-ofi-nccl/lib:/opt/amazon/efa/lib64:/usr/local/cuda-11.0/efa/lib:/usr/local/cuda-11.0/lib:/usr/local/cuda-11.0/lib64:/usr/local/cuda-11.0:/opt/nccl/build/lib:/opt/aws-ofi-nccl-inst
+#all/lib:/opt/aws-ofi-nccl/lib:$LD_LIBRARY_PATH
+export NCCL_PROTO=simple
+export PATH=/opt/amazon/efa/bin:$PATH
+export LD_PRELOAD="/opt/nccl/build/lib/libnccl.so"
+export FI_EFA_FORK_SAFE=1
+export FI_LOG_LEVEL=1
+export FI_EFA_USE_DEVICE_RDMA=1 # use for p4dn
+export NCCL_DEBUG=info
+export PYTHONFAULTHANDLER=1
+export CUDA_LAUNCH_BLOCKING=0
+export OMPI_MCA_mtl_base_verbose=1
+export FI_EFA_ENABLE_SHM_TRANSFER=0
+export FI_PROVIDER=efa
+export FI_EFA_TX_MIN_CREDITS=64
+export NCCL_TREE_THRESHOLD=0
+
+# sent to sub script
+export HOSTNAMES=`scontrol show hostnames "$SLURM_JOB_NODELIST"`
+export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
+export MASTER_PORT=12802
+export COUNT_NODE=`scontrol show hostnames "$SLURM_JOB_NODELIST" | wc -l`
+export WORLD_SIZE=$COUNT_NODE
+
+echo go $COUNT_NODE
+echo $HOSTNAMES
+echo $WORLD_SIZE
+
+mpirun -n $COUNT_NODE -perhost 1 /fsx/stable-diffusion/stable-diffusion/scripts/slurm/v1_iahr_torch111/launcher.sh

scripts/slurm/v3_pretraining/launcher.sh

@@ -16,13 +16,16 @@ echo "##########################################"
 # no magic there, just a miniconda python=3.9, pytorch=1.12, cudatoolkit=11.3
 # env with pip dependencies from stable diffusion's requirements.txt
 eval "$(/fsx/stable-diffusion/debug/miniconda3/bin/conda shell.bash hook)"
-conda activate stable
+#conda activate stable
+# torch 1.11 to avoid bug in ckpt restoring
+conda activate torch111
 cd /fsx/stable-diffusion/stable-diffusion
 
 CONFIG=configs/stable-diffusion/v3_pretraining.yaml
 
 # resume and set new seed to reshuffle data
-EXTRA="--seed 714 model.params.ckpt_path=/fsx/stable-diffusion/stable-diffusion/rlogs/2022-07-11T22-57-10_txt2img-v2-clip-encoder-improved_aesthetics-256/checkpoints/last.ckpt"
+#EXTRA="--seed 714 model.params.ckpt_path=/fsx/stable-diffusion/stable-diffusion/rlogs/2022-07-11T22-57-10_txt2img-v2-clip-encoder-improved_aesthetics-256/checkpoints/last.ckpt"
+EXTRA="--seed 715 --resume_from_checkpoint /fsx/stable-diffusion/stable-diffusion/logs/2022-07-14T21-03-49_txt2img-v2-clip-encoder-improved_aesthetics-256/checkpoints/last.ckpt"
 
 # custom logdir
 #EXTRA="${EXTRA} --logdir rlogs"
@@ -30,4 +33,7 @@ EXTRA="--seed 714 model.params.ckpt_path=/fsx/stable-diffusion/stable-diffusion/
 # debugging
 #EXTRA="${EXTRA} -d True lightning.callbacks.image_logger.params.batch_frequency=50"
 
+# detect bad gpus early on
+/bin/bash /fsx/stable-diffusion/stable-diffusion/scripts/test_gpu.sh
+
 python main.py --base $CONFIG --gpus 0,1,2,3,4,5,6,7 -t --num_nodes ${WORLD_SIZE} --scale_lr False $EXTRA

scripts/slurm/v3_pretraining/sbatch.sh

@@ -11,8 +11,7 @@
 
 module load intelmpi
 source /opt/intel/mpi/latest/env/vars.sh
-export LD_LIBRARY_PATH=/opt/aws-ofi-nccl/lib:/opt/amazon/efa/lib64:/usr/local/cuda-11.0/efa/lib:/usr/local/cuda-11.0/lib:/usr/local/cuda-11.0/lib64:/usr/local/cuda-11.0:/opt/nccl/build/lib:/opt/aws-ofi-nccl-inst
+export LD_LIBRARY_PATH=/opt/aws-ofi-nccl/lib:/opt/amazon/efa/lib64:/usr/local/cuda-11.0/efa/lib:/usr/local/cuda-11.0/lib:/usr/local/cuda-11.0/lib64:/usr/local/cuda-11.0:/opt/nccl/build/lib:/opt/aws-ofi-nccl-install/lib:/opt/aws-ofi-nccl/lib:$LD_LIBRARY_PATH
-all/lib:/opt/aws-ofi-nccl/lib:$LD_LIBRARY_PATH
 export NCCL_PROTO=simple
 export PATH=/opt/amazon/efa/bin:$PATH
 export LD_PRELOAD="/opt/nccl/build/lib/libnccl.so"