# ✨ TouchstoneGPT-7B-Instruct: A Model for Think-on-Graph 3.0 via RAG-Factory
This Hugging Face repository hosts the `TouchstoneGPT-7B-Instruct` model, an instance of a Large Language Model (LLM) based on `Qwen/Qwen2-7B-Instruct`. This model is suitable for integration within the **Think-on-Graph 3.0 (ToG-3)** framework, a novel approach to Retrieval-Augmented Generation (RAG) that enhances LLM reasoning on heterogeneous graphs. The ToG-3 framework is implemented and further detailed in the [RAG-Factory GitHub repository](https://github.com/DataArcTech/RAG-Factory).
## Paper Abstract: Think-on-Graph 3.0
Retrieval-Augmented Generation (RAG) and Graph-based RAG has become the important paradigm for enhancing Large Language Models (LLMs) with external knowledge. However, existing approaches face a fundamental trade-off. While graph-based methods are inherently dependent on high-quality graph structures, they face significant practical constraints: manually constructed knowledge graphs are prohibitively expensive to scale, while automatically extracted graphs from corpora are limited by the performance of the underlying LLM extractors, especially when using smaller, local-deployed models. This paper presents Think-on-Graph 3.0 (ToG-3), a novel framework that introduces Multi-Agent Context Evolution and Retrieval (MACER) mechanism to overcome these limitations. Our core innovation is the dynamic construction and refinement of a Chunk-Triplets-Community heterogeneous graph index, which pioneeringly incorporates a dual-evolution mechanism of Evolving Query and Evolving Sub-Graph for precise evidence retrieval. This approach addresses a critical limitation of prior Graph-based RAG methods, which typically construct a static graph index in a single pass without adapting to the actual query. A multi-agent system, comprising Constructor, Retriever, Reflector, and Responser agents, collaboratively engages in an iterative process of evidence retrieval, answer generation, sufficiency reflection, and, crucially, evolving query and subgraph. This dual-evolving multi-agent system allows ToG-3 to adaptively build a targeted graph index during reasoning, mitigating the inherent drawbacks of static, one-time graph construction and enabling deep, precise reasoning even with lightweight LLMs. Extensive experiments demonstrate that ToG-3 outperforms compared baselines on both deep and broad reasoning benchmarks, and ablation studies confirm the efficacy of the components of MACER framework.
## ✨ Features of RAG-Factory (Think-on-Graph 3.0 Implementation)
The [RAG-Factory](https://github.com/DataArcTech/RAG-Factory) framework, which implements the concepts of Think-on-Graph 3.0, provides a factory for building advanced RAG pipelines, including:
- Standard RAG implementations
- GraphRAG architectures
- Multi-modal RAG systems
Key features include:
- Modular design for easy customization
- Support for various knowledge graph backends
- Integration with multiple LLM providers
- Configurable pipeline components
## Installation (for RAG-Factory)
To set up the RAG-Factory environment, clone the repository and install dependencies:
```bash
pip install -e .
```
## Usage (for RAG-Factory)
You can run predefined RAG pipelines using the `RAG-Factory` framework:
```bash
bash run.sh naive_rag/graph_rag/mm_rag
```
or
```bash
python main.py --config examples/graphrag/config.yaml
```
For more examples and detailed configurations, please refer to the `examples/` directory in the [RAG-Factory GitHub repository](https://github.com/DataArcTech/RAG-Factory).
## Usage of TouchstoneGPT-7B-Instruct
This `TouchstoneGPT-7B-Instruct` model is a `Qwen2-7B-Instruct`-based LLM that can be used for text generation tasks, either standalone or as a component within RAG frameworks like Think-on-Graph 3.0. Below is a code snippet using the `transformers` library to load the tokenizer and model and generate content.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained(
"IDEA-FinAI/TouchstoneGPT-7B-Instruct",
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("IDEA-FinAI/TouchstoneGPT-7B-Instruct")
prompt = "What is the sentiment of the following financial post: Positive, Negative, or Neutral?
sees #Apple at $150/share in a year (+36% from today) on growing services business."
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
```
## Citation
If you find our work on Think-on-Graph 3.0 useful for your research and applications, please consider citing the paper:
```bibtex
@misc{wu2025ToG-3,
title={Think-on-Graph 3.0: Efficient and Adaptive LLM Reasoning on Heterogeneous Graphs via Multi-Agent Dual-Evolving Context Retrieval},
author={Xiaojun Wu, Cehao Yang, Xueyuan Lin, Chengjin Xu, Xuhui Jiang, Yuanliang Sun, Hui Xiong, Jia Li, Jian Guo},
year={2025},
eprint={2509.21710},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2509.21710},
}
```
