LECTOR is designed for Content-Conditional Introduction Generation (CCIG). Given the main body of a scientific paper, it first extracts a reasoning logic graph and then generates a logic-aware introduction guided by that graph and citation context.

Paper: https://arxiv.org/abs/2605.25964
Code: https://github.com/Xiao-Youth/LECTOR
Base model: Qwen/Qwen3-4B-Instruct

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Xiao-Youth/LECTOR-4B"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

For the full LECTOR prompting, rollout, and evaluation pipeline, see the project repository.

Intended Use

This model is intended for research on scientific reasoning graph extraction, logic-aware scientific writing, and content-conditional introduction generation.

Limitations

The model may produce incorrect reasoning graphs, unsupported claims, or citation errors. Generated text should be manually checked against the source paper content and references before use.

Citation

bibtex

@misc{xiao2026lector,
      title={LECTOR: Joint Optimization of Scientific Reasoning Graphs and Introduction Generation},
      author={Jiabei Xiao and Yizhou Wang and Chen Tang and Pengze Li and Wanli Ouyang and Shixiang Tang},
      year={2026},
      eprint={2605.25964},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2605.25964},
}