Commit 732db66d authored by jameskrw's avatar jameskrw
Browse files

updated large response

parent 48eb70f2

vagen/examples/sokoban/debug_multi_action_1_turn_ppo_kl.sh

+3 −3
Original line number Diff line number Diff line
@@ -31,9 +31,9 @@ python3 -m vagen.trainer.main_ppo \ 
    data.train_files=data/sokoban-text-3-step/train.parquet \

    data.val_files=data/sokoban-text-3-step/test.parquet \

    data.train_batch_size=16 \

    data.max_prompt_length=768 \

    data.max_response_length=1024 \

    data.max_trajectory_length=1024 \

    data.max_prompt_length=1024 \

    data.max_response_length=512 \

    data.max_trajectory_length=2048 \

    data.image_key=images \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-1.5B-Instruct \

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

vagen/examples/sokoban/debug_single_action_1_turn_ppo_kl_large_response.sh

+2 −2
Original line number Diff line number Diff line
@@ -31,8 +31,8 @@ python3 -m vagen.trainer.main_ppo \ 
    data.train_files=data/sokoban-text-1-step/train.parquet \

    data.val_files=data/sokoban-text-1-step/test.parquet \

    data.train_batch_size=64 \

    data.max_prompt_length=512 \

    data.max_response_length=1024 \

    data.max_prompt_length=1024 \

    data.max_response_length=512 \

    data.max_trajectory_length=2048 \

    data.image_key=images \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-1.5B-Instruct \

CRA Git | Maintained and supported by SUSTech CRA and CCSE