Loading vagen/examples/release_experiments/mask_gae_mask_loss_bi_level.sh +1 −1 Original line number Diff line number Diff line Loading @@ -22,7 +22,7 @@ python -m vagen.env.sokoban.create_dataset \ # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-vision-6-step/train.parquet \ data.val_files=data/sokoban-vision-6-step/test.parquet \ Loading vagen/examples/release_experiments/mask_gae_mask_loss_turnwise_reward_bi_level.sh +1 −1 Original line number Diff line number Diff line Loading @@ -22,7 +22,7 @@ python -m vagen.env.sokoban.create_dataset \ # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-vision-6-step/train.parquet \ data.val_files=data/sokoban-vision-6-step/test.parquet \ Loading vagen/examples/sokoban/debug_single_action_1_turn_ppo_kl_large_response_mt_gae.sh +1 −1 Original line number Diff line number Diff line Loading @@ -26,7 +26,7 @@ fi # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-text-1-step/train.parquet \ data.val_files=data/sokoban-text-1-step/test.parquet \ Loading vagen/examples/sokoban/debug_single_action_3_turns_ppo_kl_penalty_multi_turn_gae.sh +3 −3 Original line number Diff line number Diff line Loading @@ -24,7 +24,7 @@ if [ $? -ne 0 ]; then fi python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-text-3-step/train.parquet \ data.val_files=data/sokoban-text-3-step/test.parquet \ Loading Loading @@ -68,7 +68,7 @@ python3 -m vagen.trainer.main_ppo \ trainer.critic_warmup=0 \ trainer.logger=['console','wandb'] \ trainer.project_name='vagen' \ trainer.experiment_name='debug_single_action_3_turns_ppo_3B_multi_turn_gae_temp_0.7_top_p_0.95' \ trainer.experiment_name='debug_single_action_3_turns_ppo_3B_bi_level_gae_temp_0.7_top_p_0.95' \ trainer.n_gpus_per_node=4 \ trainer.nnodes=1 \ trainer.save_freq=400 \ Loading @@ -79,4 +79,4 @@ python3 -m vagen.trainer.main_ppo \ trainer.val_before_train=True \ trainer.val_generations_to_log_to_wandb=8 \ rollout_manager.n_trajectory=1 \ 2>&1 | tee debug_single_action_3_turns_ppo_3B_multi_turn_gae_temp_0.7_top_p_0.95.log 2>&1 | tee debug_single_action_3_turns_ppo_3B_bi_level_gae_temp_0.7_top_p_0.95.log vagen/examples/sokoban/debug_single_action_5_turn_ppo_kl_mt_gae.sh +3 −3 Original line number Diff line number Diff line Loading @@ -26,7 +26,7 @@ fi # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-text-5-step/train.parquet \ data.val_files=data/sokoban-text-5-step/test.parquet \ Loading Loading @@ -68,7 +68,7 @@ python3 -m vagen.trainer.main_ppo \ trainer.critic_warmup=0 \ trainer.logger=['console','wandb'] \ trainer.project_name='vagen' \ trainer.experiment_name='debug_single_action_5_turn_ppo_kl_multi_turn_gae' \ trainer.experiment_name='debug_single_action_5_turn_ppo_kl_bi_level_gae' \ trainer.n_gpus_per_node=2 \ trainer.nnodes=1 \ trainer.save_freq=100 \ Loading @@ -79,4 +79,4 @@ python3 -m vagen.trainer.main_ppo \ trainer.val_before_train=True \ trainer.val_generations_to_log_to_wandb=8 \ rollout_manager.n_trajectory=2 \ 2>&1 | tee debug_single_action_5_turn_ppo_kl_multi_turn_gae.log 2>&1 | tee debug_single_action_5_turn_ppo_kl_bi_level_gae.log Loading
vagen/examples/release_experiments/mask_gae_mask_loss_bi_level.sh +1 −1 Original line number Diff line number Diff line Loading @@ -22,7 +22,7 @@ python -m vagen.env.sokoban.create_dataset \ # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-vision-6-step/train.parquet \ data.val_files=data/sokoban-vision-6-step/test.parquet \ Loading
vagen/examples/release_experiments/mask_gae_mask_loss_turnwise_reward_bi_level.sh +1 −1 Original line number Diff line number Diff line Loading @@ -22,7 +22,7 @@ python -m vagen.env.sokoban.create_dataset \ # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-vision-6-step/train.parquet \ data.val_files=data/sokoban-vision-6-step/test.parquet \ Loading
vagen/examples/sokoban/debug_single_action_1_turn_ppo_kl_large_response_mt_gae.sh +1 −1 Original line number Diff line number Diff line Loading @@ -26,7 +26,7 @@ fi # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-text-1-step/train.parquet \ data.val_files=data/sokoban-text-1-step/test.parquet \ Loading
vagen/examples/sokoban/debug_single_action_3_turns_ppo_kl_penalty_multi_turn_gae.sh +3 −3 Original line number Diff line number Diff line Loading @@ -24,7 +24,7 @@ if [ $? -ne 0 ]; then fi python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-text-3-step/train.parquet \ data.val_files=data/sokoban-text-3-step/test.parquet \ Loading Loading @@ -68,7 +68,7 @@ python3 -m vagen.trainer.main_ppo \ trainer.critic_warmup=0 \ trainer.logger=['console','wandb'] \ trainer.project_name='vagen' \ trainer.experiment_name='debug_single_action_3_turns_ppo_3B_multi_turn_gae_temp_0.7_top_p_0.95' \ trainer.experiment_name='debug_single_action_3_turns_ppo_3B_bi_level_gae_temp_0.7_top_p_0.95' \ trainer.n_gpus_per_node=4 \ trainer.nnodes=1 \ trainer.save_freq=400 \ Loading @@ -79,4 +79,4 @@ python3 -m vagen.trainer.main_ppo \ trainer.val_before_train=True \ trainer.val_generations_to_log_to_wandb=8 \ rollout_manager.n_trajectory=1 \ 2>&1 | tee debug_single_action_3_turns_ppo_3B_multi_turn_gae_temp_0.7_top_p_0.95.log 2>&1 | tee debug_single_action_3_turns_ppo_3B_bi_level_gae_temp_0.7_top_p_0.95.log
vagen/examples/sokoban/debug_single_action_5_turn_ppo_kl_mt_gae.sh +3 −3 Original line number Diff line number Diff line Loading @@ -26,7 +26,7 @@ fi # max_trajectory_length = max_prompt_length + max_response_length python3 -m vagen.trainer.main_ppo \ algorithm.adv_estimator=multi_turn_gae \ algorithm.adv_estimator=bi_level_gae \ algorithm.high_level_gamma=0.95 \ data.train_files=data/sokoban-text-5-step/train.parquet \ data.val_files=data/sokoban-text-5-step/test.parquet \ Loading Loading @@ -68,7 +68,7 @@ python3 -m vagen.trainer.main_ppo \ trainer.critic_warmup=0 \ trainer.logger=['console','wandb'] \ trainer.project_name='vagen' \ trainer.experiment_name='debug_single_action_5_turn_ppo_kl_multi_turn_gae' \ trainer.experiment_name='debug_single_action_5_turn_ppo_kl_bi_level_gae' \ trainer.n_gpus_per_node=2 \ trainer.nnodes=1 \ trainer.save_freq=100 \ Loading @@ -79,4 +79,4 @@ python3 -m vagen.trainer.main_ppo \ trainer.val_before_train=True \ trainer.val_generations_to_log_to_wandb=8 \ rollout_manager.n_trajectory=2 \ 2>&1 | tee debug_single_action_5_turn_ppo_kl_multi_turn_gae.log 2>&1 | tee debug_single_action_5_turn_ppo_kl_bi_level_gae.log
