Commit 53fc02ca authored by jameskrw's avatar jameskrw
Browse files

added left truncation

parent 3cf7a5da

vagen/examples/release_experiments/grpo.sh

+1 −0
Original line number Diff line number Diff line
@@ -31,6 +31,7 @@ python3 -m vagen.trainer.main_ppo \ 
    data.max_response_length=128 \

    data.max_trajectory_length=1800 \

    data.image_key=images \

    data.truncation=left \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-VL-3B-Instruct \

    actor_rollout_ref.actor.optim.lr=1e-6 \

    actor_rollout_ref.model.use_remove_padding=True \

vagen/examples/release_experiments/grpo_mask_loss.sh

+1 −0
Original line number Diff line number Diff line
@@ -31,6 +31,7 @@ python3 -m vagen.trainer.main_ppo \ 
    data.max_response_length=128 \

    data.max_trajectory_length=1800 \

    data.image_key=images \

    data.truncation=left \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-VL-3B-Instruct \

    actor_rollout_ref.actor.optim.lr=1e-6 \

    actor_rollout_ref.model.use_remove_padding=True \

vagen/examples/release_experiments/mask_gae_mask_loss.sh

+1 −0
Original line number Diff line number Diff line
@@ -31,6 +31,7 @@ python3 -m vagen.trainer.main_ppo \ 
    data.max_response_length=128 \

    data.max_trajectory_length=1800 \

    data.image_key=images \

    data.truncation=left \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-VL-3B-Instruct \

    actor_rollout_ref.actor.optim.lr=1e-6 \

    actor_rollout_ref.model.use_remove_padding=True \

vagen/examples/release_experiments/mask_gae_mask_loss_bi_level.sh

+1 −0
Original line number Diff line number Diff line
@@ -30,6 +30,7 @@ python3 -m vagen.trainer.main_ppo \ 
    data.max_prompt_length=1024 \

    data.max_response_length=128 \

    data.max_trajectory_length=1800 \

    data.truncation=left \

    data.image_key=images \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-VL-3B-Instruct \

    actor_rollout_ref.actor.optim.lr=1e-6 \

vagen/examples/release_experiments/mask_gae_mask_loss_turnwise_reward.sh

+1 −0
Original line number Diff line number Diff line
@@ -31,6 +31,7 @@ python3 -m vagen.trainer.main_ppo \ 
    data.max_response_length=128 \

    data.max_trajectory_length=1800 \

    data.image_key=images \

    data.truncation=left \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-VL-3B-Instruct \

    actor_rollout_ref.actor.optim.lr=1e-6 \

    actor_rollout_ref.model.use_remove_padding=True \

CRA Git | Maintained and supported by SUSTech CRA and CCSE