--- base_model: Qwen/Qwen3-8B library_name: peft tags: [named-entity-recognition, chain-of-thought, lora] language: [en] pipeline_tag: text-generation datasets: [structured-cot-ner/pilener-teacher-cot-agreement-v1] --- # Qwen3-8B ยท Structured Reasoning for NER Part of the code, data and models of the bachelor thesis *Structuring Universal Chain-of-thought for Named Entity Recognition* (Department of Computer Science and Engineering, University of Bologna, 2026). All the repositories cited by name belong to the same organization. Code: [GreppiFilippo/structured-cot-ner](https://github.com/GreppiFilippo/structured-cot-ner). **Second phase** of the project: a teacher reasons twice on the same queries, freely and in a structured format, and students that differ only in the format of their reasoning are trained on the traces. This is the **structured** student; the free one is `qwen3-8b-free-cot-lora`. Given a text and a list of entity types (some present, some absent), the model reasons inside its `` block with three one-line fields: `TYPES` (which requested types occur), `MENTIONS` (mentions to extract and candidates rejected) and `EDGE` (ambiguous boundaries or type assignments), then answers with a JSON object that maps every requested type to the mentions copied verbatim from the text (an empty list for an absent type). ## Training - **Base model**: `Qwen/Qwen3-8B`. - **Data**: the `structured_cot` traces of `pilener-teacher-cot-agreement-v1` (train, 3,822 queries), each with its own prompt (`structured_prompt`); target: the teacher's answer (`answer`). The free and structured traces of a query lead to the same answer, so all the students see the same queries and the same targets. - **Method**: LoRA r 16, alpha 32, dropout 0.05 on the q, k, v, o, gate, up and down projections; lr 1e-4, cosine, warmup 3%; 2 epochs; batch 1 x 8 gradient accumulation; bf16 (no quantization); maximum sequence length 5,120; loss on the assistant turn only; 5% validation, best checkpoint by validation loss; seed 42. ## Evaluation Metric: micro F1 with exact, case-insensitive mention matching per type, over the requested types; reasoning length in tokens (generated before the end of the thinking block), mean per query. Queries: every gold type plus 1-3 distractors of the same domain, fixed seed (`free-vs-structured-cot-eval`). vLLM, T 0.6, top-p 0.95, top-k 20, seed 0, up to 12,288 new tokens. Base = the same model without fine-tuning, with the free prompt. **CrossNER** (full test sets) | Set | Base F1 | Free F1 | Structured F1 | Base tokens | Free tokens | Structured tokens | |---|---|---|---|---|---|---| | AI | 0.630 | 0.702 | 0.661 | 1,296 | 463 | 307 | | Literature | 0.650 | 0.724 | 0.700 | 1,431 | 536 | 394 | | Music | 0.752 | 0.785 | 0.779 | 1,511 | 566 | 393 | | Politics | 0.713 | 0.745 | 0.728 | 1,686 | 614 | 450 | | Science | 0.737 | 0.761 | 0.752 | 1,452 | 498 | 352 | | **Mean** | **0.696** | **0.743** | **0.724** | **1,475** | **535** | **379** | **Pile-NER test split** (in-domain, 2,000 queries) | Set | Base F1 | Free F1 | Structured F1 | Base tokens | Free tokens | Structured tokens | |---|---|---|---|---|---|---| | Pile-NER | 0.525 | 0.569 | 0.565 | 1,993 | 988 | 586 | **Biomedical benchmarks** (500 sentences each, 465 for JNLPBA-Rare) | Set | Base F1 | Free F1 | Structured F1 | Base tokens | Free tokens | Structured tokens | |---|---|---|---|---|---|---| | NCBI | 0.560 | 0.708 | 0.677 | 569 | 223 | 226 | | BC5CDR | 0.741 | 0.767 | 0.749 | 792 | 289 | 246 | | BC2GM | 0.478 | 0.491 | 0.400 | 697 | 249 | 243 | | BC4CHEMD | 0.509 | 0.607 | 0.588 | 840 | 298 | 322 | | AnatEM | 0.438 | 0.407 | 0.360 | 645 | 285 | 299 | | JNLPBA | 0.565 | 0.558 | 0.569 | 990 | 362 | 277 | | JNLPBA-Rare | 0.482 | 0.443 | 0.477 | 749 | 299 | 229 | | MedMentions-Rare | 0.503 | 0.563 | 0.574 | 904 | 401 | 284 | | BioRED | 0.632 | 0.720 | 0.699 | 1,176 | 470 | 311 | | **Mean** | **0.545** | **0.585** | **0.566** | **818** | **320** | **271** | All predictions and metrics: `eval-results` (`free-vs-structured-cot/` and `free-vs-structured-cot-bio/`).