RecipeMatching D4 (Qwen3-Embedding-0.6B)

One of the three models released with the paper Recipe-Matching, Not Equivalence (Ali Habibullah, Mohammad Alshiekh, Yazan Alshoibi, Salman Khan and Naeemullah Khan, 2026). Code, data and results: https://github.com/KAUST-Academy/recipe-matching-not-equivalence.

What it is. The recipe-free rung D4 of the paper's prompt ladder: Qwen3-Embedding-0.6B fine-tuned on 6,145 rows of LLM restatements of math problems written under a prompt unrelated to MathNet-Retrieve's generation prompt, positives only and no negatives (data/llm_pairs_unrelated/pairs.jsonl in the repository). Seed 42; the full training record is run_config.json in this repository, and the training command is in the GitHub README.

Results of this checkpoint (seed 42, R@1 / R@5 / R@10; the paper reports three-seed means):

Evaluation	R@1	R@5	R@10
MathNet-Retrieve easy tier (15,000 queries, 117,088 documents)	30.16	96.19	98.91
MathNet-Retrieve hard tier	0.04	23.98	59.52
Cross-language duplicates (strict)	94.15	97.20	97.20

Usage. Encode queries with the model's query prompt, the setting every number above uses.

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("KAUSTAcademy/RecipeMatching_D4_Qwen3-Embedding-0.6B")
q = model.encode(["Find all real x with x^2 = 2x."], prompt_name="query")
d = model.encode(["Determine every real solution of x^2 - 2x = 0."])
print(model.similarity(q, d))

Citation

@misc{habibullah2026recipematchingequivalence,
      title={Recipe-Matching, Not Equivalence}, 
      author={Ali Habibullah and Mohammad Alshiekh and Yazan Alshoibi and Salman Khan and Naeemullah Khan},
      year={2026},
      eprint={2609.31927},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2609.31927}, 
}