# From Reasoning LLMs to BERT: A Two-Stage Distillation Framework for Search Relevance

arXiv: https://arxiv.org/abs/2510.11056 · HTML: https://mlsanity.com/paper/2510.11056/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- Placed in Knowledge- and tool-augmented reasoning, one of 17 directions in LLM reasoning.
- Builds on 20 of its 46 references that are in the map.
- First posted on arXiv 13 Oct 2025.

## Placed in

- LLM reasoning / [Knowledge- and tool-augmented reasoning](https://mlsanity.com/lr/knowledge-and-tool-augmented-reasoning/)
- Knowledge distillation / [Knowledge distillation for LLMs](https://mlsanity.com/distill/knowledge-distillation-for-llms/)

## Builds on (most cited first, 20 of 20)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Representation Learning with Contrastive Predictive Coding](https://mlsanity.com/paper/1807.03748/) (2018)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [A Survey of Large Language Models](https://mlsanity.com/paper/2303.18223/) (2023)
- [SimCSE: Simple Contrastive Learning of Sentence Embeddings](https://mlsanity.com/paper/2104.08821/) (2021)
- [Decoupled Weight Decay Regularization](https://mlsanity.com/paper/1711.05101/) (2017)
- [TinyBERT: Distilling BERT for Natural Language Understanding](https://mlsanity.com/paper/1909.10351/) (2019)
- [M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation](https://mlsanity.com/paper/2402.03216/) (2024)
- [Passage Re-ranking with BERT](https://mlsanity.com/paper/1901.04085/) (2019)
- [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://mlsanity.com/paper/2004.02984/) (2020)
- [Improving Efficient Neural Ranking Models with Cross-Architecture Knowledge Distillation](https://mlsanity.com/paper/2010.02666/) (2020)
- [100 Days After DeepSeek-R1: A Survey on Replication Studies and More Directions for Reasoning Language Models](https://mlsanity.com/paper/2505.00551/) (2025)
- [Distilling Inter-Class Distance for Semantic Segmentation](https://mlsanity.com/paper/2205.03650/) (2022)
- [Improving Pinterest Search Relevance Using Large Language Models](https://mlsanity.com/paper/2410.17152/) (2024)
- [Knowledge Distillation based Contextual Relevance Matching for E-commerce Product Search](https://mlsanity.com/paper/2210.01701/) (2022)
- [Reward Hacking Mitigation using Verifiable Composite Rewards](https://mlsanity.com/paper/2509.15557/) (2025)
- [Best Practices for Distilling Large Language Models into BERT for Web Search Ranking](https://mlsanity.com/paper/2411.04539/) (2024)

## Built on by (most cited first, 7 of 7)

- [MixLM: High-Throughput and Effective LLM Ranking via Text-Embedding Mix-Interaction](https://mlsanity.com/paper/2512.07846/) (2025)
- [Semantic Search At LinkedIn](https://mlsanity.com/paper/2602.07309/) (2026)
- [K-CARE: Knowledge-driven Symmetrical Contextual Anchoring and Analogical Prototype Reasoning for E-commerce Relevance](https://mlsanity.com/paper/2604.25683/) (2026)
- [BRIDGE: Bridging Reasoning In Distillation Gap Elimination via Structure-Aware Masking](https://mlsanity.com/paper/2602.17686/) (2026)
- [Rationale-Guided Knowledge Distillation for Cross-Lingual Stance Detection](https://mlsanity.com/paper/2607.18693/) (2026)
- [Training Specialist Models without Reasoning Trajectories for Domain Expert Distillation](https://mlsanity.com/paper/2609.13770/) (2026)
- [KuaFu: Compressing Long User Behavior into Understanding at Billion Scale](https://mlsanity.com/paper/2609.31045/) (2026)

## Abstract (arXiv)

Query-service relevance prediction in e-commerce search systems faces strict latency requirements that prevent the direct application of Large Language Models (LLMs). To bridge this gap, we propose a two-stage reasoning distillation framework to transfer reasoning capabilities from a powerful teacher LLM to a lightweight, deployment-friendly student model. In the first stage, we address the limitations of general-purpose LLMs by constructing a domain-adapted teacher model. This is achieved through a three-step process: domain-adaptive pre-training to inject platform knowledge, supervised fine-tuning to elicit reasoning skills, and preference optimization with a multi-dimensional reward model to ensure the generation of reliable and preference-aligned reasoning paths. This teacher can then automatically annotate massive query-service pairs from search logs with both relevance labels and reasoning chains. In the second stage, to address the challenges of architectural heterogeneity in standard distillation, we introduce Contrastive Reasoning Self-Distillation (CRSD). By modeling the behavior of the same student model under ``standard'' and ``reasoning-augmented'' inputs as a teacher-student relationship, CRSD enables the lightweight model to internalize the teacher's complex decision-making mechanisms without needing the explicit reasoning path at inference. Offline evaluations and online A/B testing in the Meituan search advertising system demonstrate that our framework achieves significant improvements across multiple metrics, validating its effectiveness and practical value.