launch training

examples/train_smolvla.sh

@@ -1,28 +1,68 @@
 #!/bin/bash
 
+#SBATCH --job-name=lerobot_smolvla_test_task1_main_pretrained
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --gpus-per-node=1
+#SBATCH --mail-type=END,FAIL
+#SBATCH --output=/lustre/fswork/projects/rech/dyf/ugz83ue/logs/slurm/lerobot_smolvla_test_task1_main_pretrained.out
+###SBATCH --nodelist=jean-zay-a101
+#SBATCH --cpus-per-task=20
+###SBATCH --exclusive
+#SBATCH --time=40:00:00
+#SBATCH --mail-user=mustafa.shukor@isir.upmc.fr
+
+
+##SBATCH --partition=gpu_p2
+##SBATCH --qos=qos_gpu-t3
+###SBATCH -C v100-32g
+##SBATCH -A dyf@v100
+
+##SBATCH --partition=gpu_p5
+##SBATCH -C a100
+###SBATCH -A dyf@a100
+##SBATCH -A lqm@a100
+##SBATCH --qos=qos_gpu_a100-dev 
+##SBATCH --qos=qos_gpu_a100-t3 
+
+#SBATCH --partition=gpu_p6
+#SBATCH -C h100
+#SBATCH -A lqm@h100
+##SBATCH --qos=qos_gpu_h100-dev 
+#SBATCH --qos=qos_gpu_h100-t4 
+
+
+cd ~/lerobot_pi
+source ~/.bashrc
+source activate lerobot_main
+export LC_ALL=C
+
+rm core-*
+
+export TRANSFORMERS_CACHE=$WORK/.cache/huggingface/transformers
+export HF_HOME=$WORK/.cache/huggingface
+export DATA_DIR=$WORK/.cache/huggingface/datasets
+export HF_LEROBOT_HOME=$WORK/.cache/huggingface/lerobot
+# export LEROBOT_HOME=
+
+export HF_DATASETS_OFFLINE=1 
+export HF_HUB_OFFLINE=1
+
+export WANDB_CACHE_DIR=/lustre/fsn1/projects/rech/dyf/ugz83ue/wandb
+export WANDB_MODE=offline
+
+export TOKENIZERS_PARALLELISM=false
+
 
 cd ~/lerobot_pi
 
-source ~/miniconda3/bin/activate
+# ###### dgx
-conda activate lerobot
+# source ~/miniconda3/bin/activate
+# conda activate lerobot
 
 
 
-export WORK=/home/mustafa_shukor
+# export WORK=/home/mustafa_shukor
-# export TRANSFORMERS_CACHE=$WORK/.cache/huggingface/transformers
-# export HF_HOME=$WORK/.cache/huggingface
-# export DATA_DIR=$WORK/.cache/huggingface/datasets
-# export HF_LEROBOT_HOME=$WORK/.cache/huggingface/lerobot
-
-# export HF_DATASETS_OFFLINE=1
-# export HF_HUB_OFFLINE=1
-
-# export WANDB_CACHE_DIR=/lustre/fsn1/projects/rech/dyf/ugz83ue/wandb
-# export WANDB_MODE=offline
-
-
-## then later
-## wandb sync wandb/offline-run-*
 
 
 
@@ -35,23 +75,32 @@ POLICY=smolvla
 POLICY_NAME=smolvla
 
 
-
-
 OFFLINE_STEPS=200000
 BATCH_SIZE=64
 
 
-TASK_NAME=lerobot_${DATASET_NAME}_${POLICY_NAME}
+# TASK_NAME=lerobot_${DATASET_NAME}_${POLICY_NAME}
-
+# TRAIN_DIR=$WORK/logs/lerobot/$TASK_NAME
+# echo $TRAIN_DIR
+# rm -r $TRAIN_DIR
+# python lerobot/scripts/train.py \
+#      --policy.type=$POLICY  \
+#      --dataset.repo_id=$REPO_ID \
+#      --output_dir=$TRAIN_DIR \
+#      --batch_size=$BATCH_SIZE \
+#      --steps=$OFFLINE_STEPS
 
 
 
+TASK_NAME=lerobot_${DATASET_NAME}_${POLICY_NAME}_pretrained
 TRAIN_DIR=$WORK/logs/lerobot/$TASK_NAME
 echo $TRAIN_DIR
-
-
 rm -r $TRAIN_DIR
-CUDA_VISIBLE_DEVICES=2 python lerobot/scripts/train.py \
+POLICY_PATH=/lustre/fswork/projects/rech/dyf/ugz83ue/logs/lerobot/lerobot_so100_community_v1_v2_v3clean2_smolpi0_lr1e-4bs64steps400000gpus4freeze32_imgtoktrue_cross_attn_gap1_vlml16_causalacttrue_sa2_smolvlm2500_nobs1_expw0.75_feat2_lrvlm1e-4_trans0true_decaylr2.5e-630000_camfalse_fps3030_idlefalse/checkpoints/280000/test_smolvla/
-     --policy.type=$POLICY  \
+python lerobot/scripts/train.py \
+     --policy.path=$POLICY_PATH  \
      --dataset.repo_id=$REPO_ID \
-     --output_dir=$TRAIN_DIR
+     --output_dir=$TRAIN_DIR \
+     --batch_size=$BATCH_SIZE \
+     --steps=$OFFLINE_STEPS
+

pyproject.toml

@@ -86,6 +86,7 @@ dynamixel = ["dynamixel-sdk>=3.7.31", "pynput>=1.7.7"]
 feetech = ["feetech-servo-sdk>=1.0.0", "pynput>=1.7.7"]
 intelrealsense = ["pyrealsense2>=2.55.1.6486 ; sys_platform != 'darwin'"]
 pi0 = ["transformers>=4.48.0"]
+smolvla = ["transformers>=4.50.3", "pytest>=8.3.5", "num2words>=0.5.14", "accelerate-1.7.0"]
 pusht = ["gym-pusht>=0.1.5 ; python_version < '4.0'"]
 stretch = [
     "hello-robot-stretch-body>=0.7.27 ; python_version < '4.0' and sys_platform == 'linux'",