Loading configs/sokoban/sokoban-vision-6-step.yaml 0 → 100644 +16 −0 Original line number Diff line number Diff line seed: 42 train_path: "./data/sokoban-vision-6-step/train.parquet" test_path: "./data/sokoban-vision-6-step/test.parquet" "env1": { "env_name": "sokoban", 'env_config': { 'num_boxes': 1, 'dim_room' : [6,6], 'max_steps': 100, 'render_mode': "vision", 'min_actions_to_succeed': 5, 'max_actions_per_step': 3 }, "train_size": 10000, "test_size": 1000, } No newline at end of file vagen/examples/release_experiments/gae.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path configs/sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading vagen/examples/release_experiments/grpo.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path configs/sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading vagen/examples/release_experiments/grpo_mask_loss.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading vagen/examples/release_experiments/mask_gae.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path configs/sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading Loading
configs/sokoban/sokoban-vision-6-step.yaml 0 → 100644 +16 −0 Original line number Diff line number Diff line seed: 42 train_path: "./data/sokoban-vision-6-step/train.parquet" test_path: "./data/sokoban-vision-6-step/test.parquet" "env1": { "env_name": "sokoban", 'env_config': { 'num_boxes': 1, 'dim_room' : [6,6], 'max_steps': 100, 'render_mode': "vision", 'min_actions_to_succeed': 5, 'max_actions_per_step': 3 }, "train_size": 10000, "test_size": 1000, } No newline at end of file
vagen/examples/release_experiments/gae.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path configs/sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading
vagen/examples/release_experiments/grpo.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path configs/sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading
vagen/examples/release_experiments/grpo_mask_loss.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading
vagen/examples/release_experiments/mask_gae.sh +3 −14 Original line number Diff line number Diff line Loading @@ -4,20 +4,9 @@ set -x export VLLM_ATTENTION_BACKEND=XFORMERS export PYTHONHASHSEED=0 python -m vagen.env.sokoban.create_dataset \ --data_dir data/sokoban-vision-6-step \ --max_action_length 6 \ --dim_room 6 6 \ --num_boxes 1 \ --max_steps 100 \ --search_depth 30 \ --start_seed 0 \ --train_ratio 0.8 \ --visual_env \ --max_action_per_step 3 \ --max_action_penalty -0.1 \ --format_reward 0.5 \ --n_candidate 20000 python3 -m vagen.env.create_dataset \ --yaml_path configs/sokoban/sokoban-vision-6-step.yaml \ --force_gen # max_trajectory_length = max_prompt_length + max_response_length Loading
