IndustryLLM: Failure-Driven LLM Training for Industrial Procurement

2026-09-25

AI Summary

This study addresses the semantic gap between informal terminology, sparse attributes, and stringent engineering standards in industrial procurement, along with associated factual fragility. Building upon Qwen3.5, we propose a failure-driven domain adaptation framework that integrates national standards with technical archives. Methodologically, we design multi-genre rewriting, confidence-routed editing, and error-guided QA synthesis mechanisms, while introducing an evidence-gated three-valued logic interface to ensure output reliability. The training pipeline unifies continual pre-training, supervised fine-tuning, and a Mixture-of-Experts architecture. Experimental results demonstrate that the proposed approach improves structured query accuracy by nearly 3%, increases online Gross Merchandise Volume (GMV) by 4.25%, enhances user satisfaction by 8.3%, and reduces inference latency to 1.5 seconds, validating its effectiveness for real-world industrial applications.

Abstract

Industrial procurement requires language models to bridge informal buyer jargon, sparse marketplace attributes, and authoritative engineering standards under strict safety tolerances. We present IndustryLLM, an open-weight industrial language model trained from Qwen3.5-35B-A3B-Base (35B total parameters with ~3B activated per token, with the vision encoder frozen). Rather than relying on generic text scaling, we introduce a failure-driven adaptation recipe spanning continued pre-training (CPT) and supervised fine-tuning (SFT). CPT leverages a curated ~100B-token corpus integrating 5B tokens of national standards (e.g., GB/T) and technical archives, 10B tokens of de-identified real-world industrial transaction and inquiry records, and 60B tokens of general replay. To overcome register mismatch and factual brittleness, we systematically reconstruct an estimated 20B-token domain subset via multi-register rewriting across 10 genres and 8 writing styles, confidence-routed minimal factual editing, and error-targeted QA synthesis (resolving colloquial typos like'42-luo-mu'->42CrMo, expanding ambiguous codes like'16674'->GB/T 16674, and clarifying conflicting dimensional specs). For downstream deployment, we formalize an evidence-gated constraint-evaluation interface enforcing three-valued logic where unverified product evidence remains unknown rather than satisfied. Offline evaluations demonstrate consistent gains on procurement-query structuring (+2.97 percentage points in exact match, 95% CI [2.11, 3.86] in No-Think mode), while randomized online A/B experiments in production yield substantial improvements (+4.25% GMV, +8.3% satisfied inquiries) alongside a latency reduction from 6-7 s to 1.5 s. Model weights and configs are released at https://huggingface.co/alibaba-multimodal-industrial-ai/IndustryLLM.

Problem

• Industrial Procurement
• Language Model Adaptation
• Register Mismatch
• Factual Brittleness
• Constraint Evaluation

Innovation

• Failure-Driven Adaptation
• Industrial Procurement
• Multi-Register Rewriting
• Evidence-Gated Constraint Evaluation
• Continued Pre-Training