Resolve_issues_Task 2: Digit Recognition

May 22, 2025 by ADMIN 41 views

**Resolve Issues for Task 2: Digit Recognition**

Guide to Resolve Setup Issues for Task 2: Digit Recognition

To help others avoid common setup issues for Task 2, here’s a streamlined guide to clone repositories, install dependencies, and run the training script successfully.

1. Clone Required Repositories

Before we begin, we need to clone the required repositories. This will ensure that we have the necessary code and dependencies to complete Task 2.

git clone https://github.com/groundlight/r1_vlm.git
git clone --branch release_2025_03_06 --single-branch https://github.com/groundlight/verifiers.git
git clone --branch release_2025_03_06 --single-branch https://github.com/groundlight/trl.git

2. Set Up Virtual Environment

Next, we need to set up a virtual environment to isolate our dependencies and ensure that our project runs smoothly.

pip install uv
uv venv
source .venv/bin/activate  # Linux/Mac
# or .venv\Scripts\activate  # Windows

3. Install Dependencies

Now that we have our virtual environment set up, we can install the required dependencies.

echo "Installing dependencies..."
uv pip install --system torch==2.5.1 setuptools wheel
uv pip install --system flash-attn==2.7.3 --no-build-isolation || echo "Flash Attention failed, continuing..."

# Adjust Python version requirements
sed -i 's/requires-python = ">=3.11"/requires-python = ">=3.10"/' verifiers/pyproject.toml
sed -i 's/requires-python = ">=3.12"/requires-python = ">=3.10"/' r1_vlm/pyproject.toml

# Install TRL
cd trl
uv pip install --system -e . --no-build-isolation
cd ..

# Install Verifiers
cd verifiers
uv pip install --system -e . --no-build-isolation
cd ..

# Install additional dependencies
uv pip install --system hatchling editables

# Install r1_vlm
cd r1_vlm
uv pip install --system -e . --no-build-isolation
cd ..

echo "Installation complete!"

4. Update train.py Parameters

To optimize GPU memory usage and training settings, we need to update the train.py parameters.

sed -i 's/vllm_device="cuda:3"/vllm_device="cuda:0"/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/per_device_train_batch_size=5/per_device_train_batch_size=2/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/num_generations=15/num_generations=2/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/vllm_gpu_memory_utilization=0.8/vllm_gpu_memory_utilization=0.45/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/gradient_checkpointing = False/gradient_checkpointing = True/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/max_completion_length=512/max_completion_length=256/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/gradient_accumulation_steps=4/gradient_accumulation_steps=8/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py

5. Fix qwen_grpo_trainer.py

To fix a variable naming issue in qwen_grpo_trainer.py, we need to update the code.

sed -i '624s/completion_ids = /completion_ids_dict = /' trl/trl/trainer/qwen_grpo_trainer.py
sed -i '629a\                    completion_ids = completion_ids_dict["ids"]' trl/trl/trainer/qwen_grpo_trainer.py

6. Run Training Script

Finally, we can run the training script.

cd r1_vlm
WANDB_MODE=disabled PYTHONPATH=src CUDA_VISIBLE_DEVICES=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True python src/r1_vlm/environments/digit_recognition_env/train.py
cd ..

Notes

Ensure the virtual environment is activated before running commands.
These changes reduce GPU memory usage and fix common errors.
If issues persist, verify Python version (3.10+) and dependency compatibility.

Conclusion

Q: What are the common setup issues that can occur during Task 2?

A: Common setup issues that can occur during Task 2 include cloning repository errors, dependency installation failures, and training script execution errors.

Q: How do I clone the required repositories for Task 2?

A: To clone the required repositories, run the following commands:

git clone https://github.com/groundlight/r1_vlm.git
git clone --branch release_2025_03_06 --single-branch https://github.com/groundlight/verifiers.git
git clone --branch release_2025_03_06 --single-branch https://github.com/groundlight/trl.git

Q: What is the purpose of setting up a virtual environment for Task 2?

A: Setting up a virtual environment isolates the dependencies and ensures that the project runs smoothly. This is especially important for Task 2, as it involves complex dependencies and training scripts.

Q: How do I install the required dependencies for Task 2?

A: To install the required dependencies, run the following commands:

echo "Installing dependencies..."
uv pip install --system torch==2.5.1 setuptools wheel
uv pip install --system flash-attn==2.7.3 --no-build-isolation || echo "Flash Attention failed, continuing..."

# Adjust Python version requirements
sed -i 's/requires-python = ">=3.11"/requires-python = ">=3.10"/' verifiers/pyproject.toml
sed -i 's/requires-python = ">=3.12"/requires-python = ">=3.10"/' r1_vlm/pyproject.toml

# Install TRL
cd trl
uv pip install --system -e . --no-build-isolation
cd ..

# Install Verifiers
cd verifiers
uv pip install --system -e . --no-build-isolation
cd ..

# Install additional dependencies
uv pip install --system hatchling editables

# Install r1_vlm
cd r1_vlm
uv pip install --system -e . --no-build-isolation
cd ..

echo "Installation complete!"

Q: How do I update the train.py parameters for Task 2?

A: To update the train.py parameters, run the following commands:

sed -i 's/vllm_device="cuda:3"/vllm_device="cuda:0"/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/per_device_train_batch_size=5/per_device_train_batch_size=2/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/num_generations=15/num_generations=2/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/vllm_gpu_memory_utilization=0.8/vllm_gpu_memory_utilization=0.45/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/gradient_checkpointing = False/gradient_checkpointing = True/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/max_completion_length=512/max_completion_length=256/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py
sed -i 's/gradient_accumulation_steps=4/gradient_accumulation_steps=8/' r1_vlm/src/r1_vlm/environments/digit_recognition_env/train.py

Q: How do I fix the qwen_grpo_trainer.py issue for Task 2?

A: To fix the qwen_grpo_trainer.py issue, run the following commands:

sed -i '624s/completion_ids = /completion_ids_dict = /' trl/trl/trainer/qwen_grpo_trainer.py
sed -i '629a\                    completion_ids = completion_ids_dict["ids"]' trl/trl/trainer/qwen_grpo_trainer.py

Q: How do I run the training script for Task 2?

A: To run the training script, run the following command:

cd r1_vlm
WANDB_MODE=disabled PYTHONPATH=src CUDA_VISIBLE_DEVICES=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True python src/r1_vlm/environments/digit_recognition_env/train.py
cd ..

Q: What are the notes for Task 2?

A: The notes for Task 2 include:

Ensure the virtual environment is activated before running commands.
These changes reduce GPU memory usage and fix common errors.
If issues persist, verify Python version (3.10+) and dependency compatibility.