Qwen GRPO Graph Extraction RL Finetune

We initially explored GRPO based on the Reasoning model synthetic CoT Graph Extraction data, with LLM involved in the reward function.

.
├── LICENSE
├── ground_truth_gen # data gen via DeepSeek R1
│   ├── polished_rl_training_data.csv
│   └── r1_distill_reasoning_graph_extraction.ipynb
└── train
    └── Qwen_GRPO_Graph_Extraction.ipynb # training process

update: Seems the training notebook doesnt render properly in github, check from colab instead:

Data Gen	Training

Credits

• DeepSeek-Math and DeepSeek R1's work and DeepSeek R1 Model
  • https://arxiv.org/html/2501.12948
  • https://arxiv.org/html/2402.03300
• Qwen's Great Base model
• Will's, this(cannot find the author), and Unsloth's Daniel Han Chen work