update script

pip install bs4

pip install svgpathtools

pip install cairosvg

# To run experiment of SVG simply copy the code below

bash vagen\examples\debug_svg_vision_grpo\run_structure_only.sh

bash vagen\examples\debug_svg_vision_grpo\run_dino_only.sh

```

 No newline at end of file

Rewards:

- Overall visual similarity: +5.0

- Structural accuracy: +3.0

- Color fidelity: +2.0

- Code efficiency (using appropriate elements): +2.0

- Complete reproduction: +10.0

- Structural accuracy: +20.0

Please think step by step and provide the svg code.

Your response should be in the format of <think>...</think><answer>...</answer>

{observation}

Reward for this attempt: {reward}

Try to revise your code to make it more precise and similar to the original image.

You need to revise your code to make it more precise and similar to the original image.

Your response should be in the format of <think>...</think><answer>...</answer>

"""

 No newline at end of file

env1:

    env_name: svg

    env_config:

        split: train

    train_size: 1000

    test_size: 0

env2:

    env_name: svg

    env_config:

        split: test

    train_size: 0 

    test_size: 200

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=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=4 \

    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=0.5 \

    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_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