The value of this work lies not in proposing another isolated algorithm, but in turning enterprise knowledge injection into a systematic, verifiable, and reusable post-training pipeline: task-oriented Document-to-QA construction, SFT with general-data replay, and residual-error-focused GRPO. Wnuan improves the AAR of a 32B model from 52.76% to 91.51%, while residual-error sampling significantly outperforms full-pool and random sampling under the same update budget. Importantly, the paper also examines the associated loss in instruction-following ability and shows that RAG and model specialization are not necessarily additive. It contributes more than a high-performing model: it offers an evidence-driven methodology for deploying proprietary enterprise knowledge while balancing accuracy, training efficiency, and capability retention.</p>\n","updatedAt":"2026-08-04T17:16:18.807Z","author":{"_id":"642f6c64f945a8a5c9ee5b5d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/642f6c64f945a8a5c9ee5b5d/qz6gvAkU0xPMQ82xo5J0v.png","fullname":"XiaofengShi","name":"XiaofengAlg","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":11,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8954614996910095},"editors":["XiaofengAlg"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/642f6c64f945a8a5c9ee5b5d/qz6gvAkU0xPMQ82xo5J0v.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.01862","authors":[{"_id":"6a7151e3ec5082b9f872cd12","name":"Xiaofeng Shi","hidden":false},{"_id":"6a7151e3ec5082b9f872cd13","name":"Xiaosong Qiu","hidden":false},{"_id":"6a7151e3ec5082b9f872cd14","name":"Wenxin Ma","hidden":false},{"_id":"6a7151e3ec5082b9f872cd15","name":"Qian Kou","hidden":false},{"_id":"6a7151e3ec5082b9f872cd16","name":"Yiming Pan","hidden":false},{"_id":"6a7151e3ec5082b9f872cd17","name":"Longbin Yu","hidden":false},{"_id":"6a7151e3ec5082b9f872cd18","name":"Ying Liu","hidden":false},{"_id":"6a7151e3ec5082b9f872cd19","name":"Haiping Wang","hidden":false},{"_id":"6a7151e3ec5082b9f872cd1a","name":"Hua Zhou","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/642f6c64f945a8a5c9ee5b5d/JVDrdDZsVPH7ZZF722Bup.png","https://cdn-uploads.huggingface.co/production/uploads/642f6c64f945a8a5c9ee5b5d/9mEWh5gTE6Hhz0lQozC6z.png"],"publishedAt":"2026-08-03T00:00:00.000Z","submittedOnDailyAt":"2026-08-04T00:00:00.000Z","title":"Wnuan: Staged Post-Training for Question Answering over Proprietary Enterprise Knowledge","submittedOnDailyBy":{"_id":"642f6c64f945a8a5c9ee5b5d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/642f6c64f945a8a5c9ee5b5d/qz6gvAkU0xPMQ82xo5J0v.png","isPro":false,"fullname":"XiaofengShi","user":"XiaofengAlg","type":"user","name":"XiaofengAlg"},"summary":"Enterprise question answering requires models to acquire proprietary knowledge without discarding general capabilities. We present Wnuan, a three-stage pipeline that constructs task-oriented supervision from documents, performs supervised fine-tuning with general-data replay, and applies reinforcement learning to residual errors. On the 707-question WnuanBench, the primary 32B route raises acceptable-answer rate (AAR) from 52.76% before adaptation to 80.06% after SFT and 91.51% after RL. Under a matched 100-update protocol, residual-error sampling outperforms full-pool and size-matched random sampling by 3.11 and 2.97 points, respectively. Source-cluster bootstrap intervals remain above zero for both contrasts, and a same-domain validation set preserves the ordering. The general-benchmark average decreases by 5.17 points across the route, concentrated in instruction following. The automatic evaluation ensemble agrees with an authoritative domain expert on 90.5% of a stratified Wnuan-Inst response sample. These results characterize both the gains and the general-capability cost of staged enterprise adaptation.","upvotes":3,"discussionId":"6a7151e3ec5082b9f872cd1b","organization":{"_id":"61be9739d2f9358e24ca0a4f","name":"BAAI","fullname":"Beijing Academy of Artificial Intelligence","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1664511063789-632c234f42c386ebd2710434.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"642f6c64f945a8a5c9ee5b5d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/642f6c64f945a8a5c9ee5b5d/qz6gvAkU0xPMQ82xo5J0v.png","isPro":false,"fullname":"XiaofengShi","user":"XiaofengAlg","type":"user"},{"_id":"63c1699e40a26dd2db32400d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63c1699e40a26dd2db32400d/3N0-Zp8igv8-52mXAdiiq.jpeg","isPro":false,"fullname":"Chroma","user":"Chroma111","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"61be9739d2f9358e24ca0a4f","name":"BAAI","fullname":"Beijing Academy of Artificial Intelligence","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1664511063789-632c234f42c386ebd2710434.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.01862.md","query":{}}">
Wnuan: Staged Post-Training for Question Answering over Proprietary Enterprise Knowledge
Abstract
Enterprise question answering requires models to acquire proprietary knowledge without discarding general capabilities. We present Wnuan, a three-stage pipeline that constructs task-oriented supervision from documents, performs supervised fine-tuning with general-data replay, and applies reinforcement learning to residual errors. On the 707-question WnuanBench, the primary 32B route raises acceptable-answer rate (AAR) from 52.76% before adaptation to 80.06% after SFT and 91.51% after RL. Under a matched 100-update protocol, residual-error sampling outperforms full-pool and size-matched random sampling by 3.11 and 2.97 points, respectively. Source-cluster bootstrap intervals remain above zero for both contrasts, and a same-domain validation set preserves the ordering. The general-benchmark average decreases by 5.17 points across the route, concentrated in instruction following. The automatic evaluation ensemble agrees with an authoritative domain expert on 90.5% of a stratified Wnuan-Inst response sample. These results characterize both the gains and the general-capability cost of staged enterprise adaptation.
Community
The value of this work lies not in proposing another isolated algorithm, but in turning enterprise knowledge injection into a systematic, verifiable, and reusable post-training pipeline: task-oriented Document-to-QA construction, SFT with general-data replay, and residual-error-focused GRPO. Wnuan improves the AAR of a 32B model from 52.76% to 91.51%, while residual-error sampling significantly outperforms full-pool and random sampling under the same update budget. Importantly, the paper also examines the associated loss in instruction-following ability and shows that RAG and model specialization are not necessarily additive. It contributes more than a high-performing model: it offers an evidence-driven methodology for deploying proprietary enterprise knowledge while balancing accuracy, training efficiency, and capability retention.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.01862 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.01862 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.01862 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.