Commit 748dda26 authored by YaningGao's avatar YaningGao
Browse files

add svg script

parent 66153008

vagen/examples/debug_frozen_lake_text_grpo_service/run.sh

100644 → 100755
+3 −3
Original line number Diff line number Diff line
@@ -21,7 +21,7 @@ python3 -m vagen.trainer.main_ppo \ 
    data.train_batch_size=16 \

    data.max_prompt_length=1024 \

    data.max_response_length=128 \

    data.max_trajectory_length=1800 \

    data.max_trajectory_length=3600 \

    data.image_key=images \

    data.truncation=error \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-0.5B-Instruct \
@@ -46,7 +46,7 @@ python3 -m vagen.trainer.main_ppo \ 
    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 \

    actor_rollout_ref.rollout.temperature=0.5 \

    critic.optim.lr=1e-5 \

    critic.model.use_remove_padding=True \

    critic.model.path=Qwen/Qwen2.5-VL-3B-Instruct \
@@ -54,7 +54,7 @@ python3 -m vagen.trainer.main_ppo \ 
    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 \

    algorithm.kl_ctrl.kl_coef=0.01 \

    trainer.critic_warmup=0 \

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

    trainer.project_name='vagen_debug' \

vagen/examples/debug_svg_vision_grpo/env_config.yaml→vagen/examples/svg_aico/env_config.yaml

+2 −2
Original line number Diff line number Diff line
@@ -2,7 +2,7 @@ env1: 
    env_name: svg

    env_config:

        split: train

        dino_weight: 0.0

        dino_weight: 1.0

        structural_weight: 1.0

    train_size: 10000

    test_size: 0
@@ -11,7 +11,7 @@ env2: 
    env_name: svg

    env_config:

        split: test

        dino_weight: 0.0

        dino_weight: 1.0

        structural_weight: 1.0

    train_size: 0 

    test_size: 512

vagen/examples/svg_aico/run.sh

0 → 100644
+75 −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.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=masked_gae \

    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=1672 \

    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=32 \

    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=0.2 \

    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_new' \

    trainer.experiment_name='aico_svg_vision' \

    trainer.n_gpus_per_node=4 \

    trainer.nnodes=1 \

    trainer.save_freq=70 \

    trainer.test_freq=20 \

    trainer.total_training_steps=200 \

    rollout_manager.max_turns=3 \

    rollout_manager.window_size=5 \

    rollout_manager.use_multi_turn_reward=False \

    rollout_manager.use_loss_mask=True \

    rollout_manager.use_gae_mask=True \

    trainer.val_before_train=True \

    trainer.val_generations_to_log_to_wandb=8 \

    rollout_manager.n_trajectory=8 \

    2>&1 | tee aico_frozenlake_vision.log
 
 No newline at end of file

vagen/examples/svg_trico/env_config.yaml

0 → 100644
+19 −0
Original line number Diff line number Diff line 
env1:

    env_name: svg

    env_config:

        split: train

        dino_weight: 1.0

        structural_weight: 1.0

    train_size: 10000

    test_size: 0



env2:

    env_name: svg

    env_config:

        split: test

        dino_weight: 1.0

        structural_weight: 1.0

    train_size: 0 

    test_size: 512

vagen/examples/svg_trico/run.sh

0 → 100644
+75 −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.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=bi_level_gae \

    algorithm.high_level_gamma=1.0 \

    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=32 \

    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=0.2 \

    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_new' \

    trainer.experiment_name='trico_svg_vision' \

    trainer.n_gpus_per_node=4 \

    trainer.nnodes=1 \

    trainer.save_freq=70 \

    trainer.test_freq=20 \

    trainer.total_training_steps=200 \

    rollout_manager.max_turns=3 \

    rollout_manager.window_size=5 \

    rollout_manager.use_multi_turn_reward=False \

    rollout_manager.use_loss_mask=True \

    rollout_manager.use_gae_mask=True \

    trainer.val_before_train=True \

    trainer.val_generations_to_log_to_wandb=8 \

    rollout_manager.n_trajectory=8 \

    2>&1 | tee trico_frozenlake_vision.log
 
 No newline at end of file

CRA Git | Maintained and supported by SUSTech CRA and CCSE