# ChemDFM-R: A Chemical Reasoning LLM Enhanced with Atomized Chemical Knowledge While large language models (LLMs) have achieved impressive progress, their application in scientific domains such as chemistry remains hindered by shallow domain understanding and limited reasoning capabilities. In this work, we focus on the specific field of chemistry and develop a Chemical Reasoning LLM, **ChemDFM-R**. We first construct a comprehensive dataset of atomized chemical knowledge, *ChemFG*, annotating the presence of functional groups in molecules and the changes of functional groups during chemical reactions, to enhance the model’s understanding of the fundamental principles and internal logic of chemistry. Then, we propose a *mix-sourced distillation* method that integrates expertise in atomized knowledge with general reasoning skills, followed by domain-specific reinforcement learning to enhance chemical reasoning. Experiments on diverse chemical benchmarks demonstrate that ChemDFM-R achieves cutting-edge performance while providing interpretable, rationale-driven outputs. Further case studies illustrate how explicit reasoning chains significantly improve the model's reliability, transparency, and practicality in real-world human-AI collaboration scenarios. For more details, please refer to [our paper](https://arxiv.org/abs/2507.21990). ## News * 🎉 **2025-12-29**: The functional group identification toolkit, ChemFG-Tool, is now open-sourced on [GitHub](https://github.com/OpenDFM/ChemFG-Tool)! * **2025-10-26**: The parameter of [ChemDFM-R-14B](https://huggingface.co/OpenDFM/ChemDFM-R-14B) is open-sourced! * **2025-10-26**: [ChemDFM-v2.0-14B](https://huggingface.co/OpenDFM/ChemDFM-v2.0-14B) is released! The improved domain pre-training and instruction tuning procedure is implemented on Qwen2.5-14B to achieve a more advanced general LLM in Chemistry. More details can be found [here](https://huggingface.co/OpenDFM/ChemDFM-v2.0-14B). * **2025-07-29**: The paper of ChemDFM-R-14B is released on arXiv: [ChemDFM-R: A Chemical Reasoning LLM Enhanced with Atomized Chemical Knowledge](https://arxiv.org/abs/2507.21990). * 2024-11-09: [ChemDFM-v1.5-8B](https://huggingface.co/OpenDFM/ChemDFM-v1.5-8B) is released! We implemented our domain pre-training and instruction tuning procedure on a stronger base model LLaMA-3-8B. * 2024-03-12: The parameter of [ChemDFM-v1.0-13B](https://huggingface.co/OpenDFM/ChemDFM-v1.0-13B) is open-sourced! * 2024-01-26: The paper of ChemDFM-13B is released on arXiv: [ChemDFM: Dialogue Foundation Model for Chemistry](https://arxiv.org/abs/2401.14818) ### local inference To load and run ChemDFM-R locally, here is an example: ```python import re import torch from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig model_name_or_id = "OpenDFM/ChemDFM-R-14B" tokenizer = AutoTokenizer.from_pretrained(model_name_or_id) model = AutoModelForCausalLM.from_pretrained(model_name_or_id, torch_dtype=torch.float16).to("cuda") instruction = "Can you please give detailed descriptions of the molecule below?\nCl.O=C1c2c(O)cccc2-c2nn(CCNCCO)c3ccc(NCCNCCO)c1c23" message = [ { "role": "system", "content": "You are a helpful assistant that is good at reasoning. You always reason thoroughly before giving response. The reasoning process and answer are enclosed within and tags, respectively.\ni.e.,\n\nreasoning process here\n\n\nanswer here\n" }, { "role": "user", "content": instruction } ] input_text = tokenizer.apply_chat_template(message, tokenize=False, add_generation_prompt=True) inputs = tokenizer(input_text, return_tensors="pt").to("cuda") generation_config = GenerationConfig( do_sample=True, top_k=20, top_p=0.9, temperature=0.9, max_new_tokens=1024, repetition_penalty=1.05, eos_token_id=tokenizer.eos_token_id ) outputs = model.generate(**inputs, generation_config=generation_config) generated_text = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0] input_text = tokenizer.decode(inputs["input_ids"][0], skip_special_tokens=True) generated_text = generated_text[len(input_text):].strip() print(f"{generated_text=}") thinking, answer = re.match(r'(.*?)\s?(.*?)', generated_text, re.DOTALL).groups() thinking, answer = thinking.strip(), answer.strip() print(f"{thinking=}") print(f"{answer=}") ``` ### SMILES preprocess When there involves SMILES notation in your input, we recommend to preprocess the SMILES with the `rdkit` package to canonicalize the SMILES. Here is an example: ```python from rdkit import Chem def canonicalize_smiles(smiles): mol = Chem.MolFromSmiles(smiles) if mol is None: return None return Chem.MolToSmiles(mol, isomericSmiles=True, kekuleSmiles=False) ``` or directly: ```python from rdkit import Chem def canonicalize_smiles(smiles): return Chem.CanonSmiles(smiles, useChiral=True) ``` ## Citation ```bibtex @misc{zhao2025chemdfmr, title={ChemDFM-R: An Chemical Reasoner LLM Enhanced with Atomized Chemical Knowledge}, author={Zihan Zhao and Bo Chen and Ziping Wan and Lu Chen and Xuanze Lin and Shiyang Yu and Situo Zhang and Da Ma and Zichen Zhu and Danyang Zhang and Huayang Wang and Zhongyang Dai and Liyang Wen and Xin Chen and Kai Yu}, year={2025}, eprint={2507.21990}, archivePrefix={arXiv}, primaryClass={cs.CE}, url={https://arxiv.org/abs/2507.21990}, } ``` ## Disclaimer Current version of ChemDFM may generate incorrect or misleading information. Please use it with caution and verify the results with domain experts before making any decisions based on the results. ## Contact If you have any questions or further requests, please contact [Zihan Zhao](mailto:zhao_mengxin@sjtu.edu.cn), [Bo Chen](mailto:chenb@szlab.ac.cn), and [Lu Chen](mailto:chenlusz@sjtu.edu.cn).