Commit f869a32d authored by YaningDylan's avatar YaningDylan
Browse files

update script

parent 9a205844

tmpn40k8k1n

deleted100644 → 0
+0 −172
Original line number Diff line number Diff line 


Section "Device"

    Identifier     "Device0"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:5:0:0"

EndSection





Section "Screen"

    Identifier     "Screen0"

    Device         "Device0"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "Device"

    Identifier     "Device1"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:6:0:0"

EndSection





Section "Screen"

    Identifier     "Screen1"

    Device         "Device1"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "Device"

    Identifier     "Device2"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:9:0:0"

EndSection





Section "Screen"

    Identifier     "Screen2"

    Device         "Device2"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "Device"

    Identifier     "Device3"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:10:0:0"

EndSection





Section "Screen"

    Identifier     "Screen3"

    Device         "Device3"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "Device"

    Identifier     "Device4"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:141:0:0"

EndSection





Section "Screen"

    Identifier     "Screen4"

    Device         "Device4"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "Device"

    Identifier     "Device5"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:142:0:0"

EndSection





Section "Screen"

    Identifier     "Screen5"

    Device         "Device5"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "Device"

    Identifier     "Device6"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:145:0:0"

EndSection





Section "Screen"

    Identifier     "Screen6"

    Device         "Device6"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "Device"

    Identifier     "Device7"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BusID          "PCI:146:0:0"

EndSection





Section "Screen"

    Identifier     "Screen7"

    Device         "Device7"

    DefaultDepth    24

    Option         "AllowEmptyInitialConfiguration" "True"

    SubSection     "Display"

        Depth       24

        Virtual 1024 768

    EndSubSection

EndSection





Section "ServerLayout"

    Identifier     "Layout0"

    Screen 0 "Screen0" 0 0

    Screen 1 "Screen1" 0 0

    Screen 2 "Screen2" 0 0

    Screen 3 "Screen3" 0 0

    Screen 4 "Screen4" 0 0

    Screen 5 "Screen5" 0 0

    Screen 6 "Screen6" 0 0

    Screen 7 "Screen7" 0 0

EndSection

vagen/examples/debug_svg_vision_grpo/env_config_dino.yaml

0 → 100644
+17 −0
Original line number Diff line number Diff line 
env1:

    env_name: svg

    env_config:

        split: train

        dino_only: True

    train_size: 10000

    test_size: 0



env2:

    env_name: svg

    env_config:

        split: test

        dino_only: True

    train_size: 0 

    test_size: 512

vagen/examples/debug_svg_vision_grpo/env_config_structure.yaml

0 → 100644
+23 −0
Original line number Diff line number Diff line 
env1:

    env_name: svg

    env_config:

        split: train

        dino_weight: 0.0

        structural_weight: 1.0

        color_weight: 0.0

        code_weight: 0.0

    train_size: 10000

    test_size: 0



env2:

    env_name: svg

    env_config:

        split: test

        dino_weight: 0.0

        structural_weight: 1.0

        color_weight: 0.0

        code_weight: 0.0

    train_size: 0 

    test_size: 512

vagen/examples/debug_svg_vision_grpo/run_dino_only.sh

0 → 100644
+74 −0
Original line number Diff line number Diff line 
set -x





export VLLM_ATTENTION_BACKEND=XFORMERS

export PYTHONHASHSEED=0



SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"



python -m vagen.env.create_dataset \

    --yaml_path "$SCRIPT_DIR/env_config_dino.yaml" \

    --train_path "data/svg-vision-debug/train.parquet" \

    --test_path "data/svg-vision-debug/test.parquet" \



# max_trajectory_length = max_prompt_length + max_response_length



python3 -m vagen.trainer.main_ppo \

    algorithm.adv_estimator=grpo \

    algorithm.high_level_gamma=0.95 \

    data.train_files=data/svg-vision-debug/train.parquet \

    data.val_files=data/svg-vision-debug/test.parquet \

    data.train_batch_size=16 \

    data.max_prompt_length=1024 \

    data.max_response_length=648 \

    data.max_trajectory_length=3600 \

    data.image_key=images \

    data.truncation=error \

    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 \

    actor_rollout_ref.actor.ppo_mini_batch_size=8 \

    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=1 \

    actor_rollout_ref.actor.use_kl_loss=False \

    actor_rollout_ref.actor.kl_loss_coef=0.001 \

    actor_rollout_ref.actor.kl_loss_type=mse \

    actor_rollout_ref.model.enable_gradient_checkpointing=True \

    actor_rollout_ref.actor.fsdp_config.param_offload=False \

    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \

    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=1 \

    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \

    actor_rollout_ref.rollout.name=vllm \

    actor_rollout_ref.rollout.gpu_memory_utilization=0.3 \

    actor_rollout_ref.rollout.enable_chunked_prefill=False \

    actor_rollout_ref.rollout.enforce_eager=False \

    actor_rollout_ref.rollout.free_cache_engine=False \

    actor_rollout_ref.rollout.n=1 \

    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=1 \

    actor_rollout_ref.ref.fsdp_config.param_offload=True \

    actor_rollout_ref.rollout.top_p=0.95 \

    actor_rollout_ref.rollout.temperature=0.7 \

    critic.optim.lr=1e-5 \

    critic.model.use_remove_padding=True \

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

    critic.model.enable_gradient_checkpointing=True \

    critic.ppo_micro_batch_size_per_gpu=1 \

    critic.model.fsdp_config.param_offload=False \

    critic.model.fsdp_config.optimizer_offload=False \

    algorithm.kl_ctrl.kl_coef=0.001 \

    trainer.critic_warmup=0 \

    trainer.logger=['console','wandb'] \

    trainer.project_name='vagen_debug' \

    trainer.experiment_name='grpo_mask_loss_dino_svg_vision_debug' \

    trainer.n_gpus_per_node=2 \

    trainer.nnodes=1 \

    trainer.save_freq=100 \

    trainer.test_freq=20 \

    trainer.total_training_steps=300 \

    rollout_manager.max_turns=3 \

    rollout_manager.window_size=5 \

    rollout_manager.use_multi_turn_reward=False \

    rollout_manager.use_loss_mask=True \

    trainer.val_before_train=True \

    trainer.val_generations_to_log_to_wandb=8 \

    rollout_manager.n_trajectory=8 \

    2>&1 | tee grpo_mask_loss_svg_vision_debug.log

vagen/examples/debug_svg_vision_grpo/run_structure_only.sh

0 → 100644
+74 −0
Original line number Diff line number Diff line 
set -x





export VLLM_ATTENTION_BACKEND=XFORMERS

export PYTHONHASHSEED=0



SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"



python -m vagen.env.create_dataset \

    --yaml_path "$SCRIPT_DIR/env_config_structure.yaml" \

    --train_path "data/svg-vision-debug/train.parquet" \

    --test_path "data/svg-vision-debug/test.parquet" \



# max_trajectory_length = max_prompt_length + max_response_length



python3 -m vagen.trainer.main_ppo \

    algorithm.adv_estimator=grpo \

    algorithm.high_level_gamma=0.95 \

    data.train_files=data/svg-vision-debug/train.parquet \

    data.val_files=data/svg-vision-debug/test.parquet \

    data.train_batch_size=16 \

    data.max_prompt_length=1024 \

    data.max_response_length=648 \

    data.max_trajectory_length=3600 \

    data.image_key=images \

    data.truncation=error \

    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 \

    actor_rollout_ref.actor.ppo_mini_batch_size=8 \

    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=1 \

    actor_rollout_ref.actor.use_kl_loss=False \

    actor_rollout_ref.actor.kl_loss_coef=0.001 \

    actor_rollout_ref.actor.kl_loss_type=mse \

    actor_rollout_ref.model.enable_gradient_checkpointing=True \

    actor_rollout_ref.actor.fsdp_config.param_offload=False \

    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \

    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=1 \

    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \

    actor_rollout_ref.rollout.name=vllm \

    actor_rollout_ref.rollout.gpu_memory_utilization=0.3 \

    actor_rollout_ref.rollout.enable_chunked_prefill=False \

    actor_rollout_ref.rollout.enforce_eager=False \

    actor_rollout_ref.rollout.free_cache_engine=False \

    actor_rollout_ref.rollout.n=1 \

    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=1 \

    actor_rollout_ref.ref.fsdp_config.param_offload=True \

    actor_rollout_ref.rollout.top_p=0.95 \

    actor_rollout_ref.rollout.temperature=0.7 \

    critic.optim.lr=1e-5 \

    critic.model.use_remove_padding=True \

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

    critic.model.enable_gradient_checkpointing=True \

    critic.ppo_micro_batch_size_per_gpu=1 \

    critic.model.fsdp_config.param_offload=False \

    critic.model.fsdp_config.optimizer_offload=False \

    algorithm.kl_ctrl.kl_coef=0.001 \

    trainer.critic_warmup=0 \

    trainer.logger=['console','wandb'] \

    trainer.project_name='vagen_debug' \

    trainer.experiment_name='grpo_mask_loss_structure_svg_vision_debug' \

    trainer.n_gpus_per_node=2 \

    trainer.nnodes=1 \

    trainer.save_freq=100 \

    trainer.test_freq=20 \

    trainer.total_training_steps=300 \

    rollout_manager.max_turns=3 \

    rollout_manager.window_size=5 \

    rollout_manager.use_multi_turn_reward=False \

    rollout_manager.use_loss_mask=True \

    trainer.val_before_train=True \

    trainer.val_generations_to_log_to_wandb=8 \

    rollout_manager.n_trajectory=8 \

    2>&1 | tee grpo_mask_loss_svg_vision_debug.log

CRA Git | Maintained and supported by SUSTech CRA and CCSE