Fine-tuned from
Qwen3.5-27B using
Self-Graph Reasoning (SGR) +
Logical Curriculum Learning (LCL) — a 4-phase progressive training approach based on LoCM (Logical Complexity Metric).
SGR is a graph-structured reasoning framework that enables LLMs to explicitly represent their reasoning processes as structured graphs prior to producing final answers. See
arXiv:2601.03597.
Standard SGR training data has a mean LoCM of 23.33 with zero samples below LoCM 6, causing the model to skip foundational logic. LCL addresses this with 4-phase progressive training:
Each phase warm-starts from the previous phase's best LoRA checkpoint. LoCM (Logical Complexity Metric) is from
arXiv:2601.02902.
ProverQA and NSA-LR evaluation datasets have
zero overlap with training data. Full audit available in the
project repository.
1from transformers import AutoModelForCausalLM, AutoTokenizer
2
3model = AutoModelForCausalLM.from_pretrained(
4 "jeffchanpm/Qwen3.5-27B-SGR-LCL",
5 torch_dtype="auto",
6 device_map="auto",
7)
8tokenizer = AutoTokenizer.from_pretrained("jeffchanpm/Qwen3.5-27B-SGR-LCL")
9
10prompt = "If all dogs are animals and all animals are living things, are all dogs living things? Think step by step."
11messages = [{"role": "user", "content": prompt}]
12text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
13inputs = tokenizer(text, return_tensors="pt").to(model.device)
14outputs = model.generate(**inputs, max_new_tokens=2048)
15print(tokenizer.decode(outputs[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True))
For GGUF quantized versions (Q4_K_M), see
jeffchanpm/Qwen3.5-27B-SGR-LCL-GGUF.
1@article{chen2026chains,
2 title={From Chains to Graphs: Self-Structured Reasoning for General-Domain LLMs},
3 author={Chen, Yingjian and Liu, Haoran and Liu, Yinhong and Tong, Sherry T and Feng, Aosong and Lu, Jinghui and Zhang, Juntao and Iwasawa, Yusuke and Matsuo, Yutaka and Li, Irene},
4 journal={arXiv preprint arXiv:2601.03597},
5 year={2026}
6}
