https://github.com/deep-diver/llamaduo

\n","updatedAt":"2024-08-27T03:24:34.529Z","author":{"_id":"60f1abe7544c2adfd699860c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg","fullname":"AK","name":"akhaliq","type":"user","isPro":false,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":9179,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6352326273918152},"editors":["akhaliq"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg"],"reactions":[{"reaction":"❤️","users":["chansung","sayakpaul","AdinaY","juyongjiang","ddiddi","curt-park","elsatch","Kat380"],"count":8}],"isReport":false}},{"id":"66ce7e5ebe6193e8ed0b9671","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false},"createdAt":"2024-08-28T01:33:18.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [CodeACT: Code Adaptive Compute-efficient Tuning Framework for Code LLMs](https://huggingface.co/papers/2408.02193) (2024)\n* [XMainframe: A Large Language Model for Mainframe Modernization](https://huggingface.co/papers/2408.04660) (2024)\n* [FANNO: Augmenting High-Quality Instruction Data with Open-Sourced LLMs Only](https://huggingface.co/papers/2408.01323) (2024)\n* [Enhanced Fine-Tuning of Lightweight Domain-Specific Q&A Model Based on Large Language Models](https://huggingface.co/papers/2408.12247) (2024)\n* [AnyTaskTune: Advanced Domain-Specific Solutions through Task-Fine-Tuning](https://huggingface.co/papers/2407.07094) (2024)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend

\n","updatedAt":"2024-08-28T01:33:18.166Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.749397337436676},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}},{"id":"670e415485a1548f5c00a22f","author":{"_id":"66bc5bcd0efef70ff356f356","avatarUrl":"/avatars/e2186d1a891e1438d52f0d8b9d441a9c.svg","fullname":"Ragulkumar R","name":"Ragulkumar1104","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2024-10-15T10:17:56.000Z","type":"comment","data":{"edited":true,"hidden":false,"latest":{"raw":"Nice","html":"

Nice

\n","updatedAt":"2024-10-15T10:18:26.128Z","author":{"_id":"66bc5bcd0efef70ff356f356","avatarUrl":"/avatars/e2186d1a891e1438d52f0d8b9d441a9c.svg","fullname":"Ragulkumar R","name":"Ragulkumar1104","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":1,"editors":["Ragulkumar1104"],"editorAvatarUrls":["/avatars/e2186d1a891e1438d52f0d8b9d441a9c.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2408.13467","authors":[{"_id":"66cd31e4a9107693c2dbafbb","user":{"_id":"60d3b57ad7b174177faabd6e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1659971187637-60d3b57ad7b174177faabd6e.jpeg","isPro":true,"fullname":"chansung park","user":"chansung","type":"user"},"name":"Chansung Park","status":"claimed_verified","statusLastChangedAt":"2024-08-27T07:38:48.610Z","hidden":false},{"_id":"66cd31e4a9107693c2dbafbc","user":{"_id":"64b63712b3b69063d9aff341","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b63712b3b69063d9aff341/nP6GyBKBtnWWoA0TvQF3K.jpeg","isPro":false,"fullname":"John","user":"juyongjiang","type":"user"},"name":"Juyong Jiang","status":"admin_assigned","statusLastChangedAt":"2024-08-27T12:48:26.456Z","hidden":false},{"_id":"66cd31e4a9107693c2dbafbd","name":"Fan Wang","hidden":false},{"_id":"66cd31e4a9107693c2dbafbe","user":{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","isPro":true,"fullname":"Sayak Paul","user":"sayakpaul","type":"user"},"name":"Sayak Paul","status":"admin_assigned","statusLastChangedAt":"2024-08-27T12:50:17.512Z","hidden":false},{"_id":"66cd31e4a9107693c2dbafbf","name":"Jing Tang","hidden":false},{"_id":"66cd31e4a9107693c2dbafc0","name":"Sunghun Kim","hidden":false}],"publishedAt":"2024-08-24T05:03:08.000Z","submittedOnDailyAt":"2024-08-27T01:54:34.514Z","title":"LlamaDuo: LLMOps Pipeline for Seamless Migration from Service LLMs to\n Small-Scale Local LLMs","submittedOnDailyBy":{"_id":"60f1abe7544c2adfd699860c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg","isPro":false,"fullname":"AK","user":"akhaliq","type":"user"},"summary":"The widespread adoption of cloud-based proprietary large language models\n(LLMs) has introduced significant challenges, including operational\ndependencies, privacy concerns, and the necessity of continuous internet\nconnectivity. In this work, we introduce an LLMOps pipeline, \"LlamaDuo\", for\nthe seamless migration of knowledge and abilities from service-oriented LLMs to\nsmaller, locally manageable models. This pipeline is crucial for ensuring\nservice continuity in the presence of operational failures, strict privacy\npolicies, or offline requirements. Our LlamaDuo involves fine-tuning a small\nlanguage model against the service LLM using a synthetic dataset generated by\nthe latter. If the performance of the fine-tuned model falls short of\nexpectations, it is enhanced by further fine-tuning with additional similar\ndata created by the service LLM. This iterative process guarantees that the\nsmaller model can eventually match or even surpass the service LLM's\ncapabilities in specific downstream tasks, offering a practical and scalable\nsolution for managing AI deployments in constrained environments. Extensive\nexperiments with leading edge LLMs are conducted to demonstrate the\neffectiveness, adaptability, and affordability of LlamaDuo across various\ndownstream tasks. Our pipeline implementation is available at\nhttps://github.com/deep-diver/llamaduo.","upvotes":25,"discussionId":"66cd31e5a9107693c2dbb010","githubRepo":"https://github.com/deep-diver/llamaduo","githubRepoAddedBy":"auto","ai_summary":"The LlamaDuo pipeline enables the transfer of knowledge from cloud-based large language models to smaller, locally manageable models through iterative fine-tuning with synthetic datasets.","ai_keywords":["LLMs","LLMOps","fine-tuning","synthetic dataset","服务质量 continuity","privacy policies","offline requirements"],"githubStars":315},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"65ba471ad88a65abb9328ee2","avatarUrl":"/avatars/956238ce5034091e64d026b0272c4400.svg","isPro":false,"fullname":"Dazhi Jiang","user":"thuzhizhi","type":"user"},{"_id":"60d3b57ad7b174177faabd6e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1659971187637-60d3b57ad7b174177faabd6e.jpeg","isPro":true,"fullname":"chansung park","user":"chansung","type":"user"},{"_id":"65b9dee19c4955ae7aee4954","avatarUrl":"/avatars/c519e2f933e2be184e96e8ebba73591c.svg","isPro":false,"fullname":"Jemin Lee","user":"leejaymin","type":"user"},{"_id":"6417ebef1f1f3b0fa8119af1","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1679289304697-noauth.jpeg","isPro":false,"fullname":"Dongjun Min","user":"petanerd","type":"user"},{"_id":"648eb1eb59c4e5c87dc116e0","avatarUrl":"/avatars/c636cea39c2c0937f01398c94ead5dad.svg","isPro":false,"fullname":"fdsqefsgergd","user":"T-representer","type":"user"},{"_id":"5ee3a7cd2a3eae3cbdad1305","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1594144055859-5ee3a7cd2a3eae3cbdad1305.jpeg","isPro":true,"fullname":"Yacine Jernite","user":"yjernite","type":"user"},{"_id":"6270324ebecab9e2dcf245de","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6270324ebecab9e2dcf245de/cMbtWSasyNlYc9hvsEEzt.jpeg","isPro":false,"fullname":"Kye Gomez","user":"kye","type":"user"},{"_id":"6640bbd0220cfa8cbfdce080","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6640bbd0220cfa8cbfdce080/wiAHUu5ewawyipNs0YFBR.png","isPro":true,"fullname":"John Smith","user":"John6666","type":"user"},{"_id":"64b63712b3b69063d9aff341","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b63712b3b69063d9aff341/nP6GyBKBtnWWoA0TvQF3K.jpeg","isPro":false,"fullname":"John","user":"juyongjiang","type":"user"},{"_id":"65226ef8ef81ce01a471f69a","avatarUrl":"/avatars/4265d691bfa9b7c1c98516772efe4376.svg","isPro":false,"fullname":"Wong","user":"Kat380","type":"user"},{"_id":"651e59c5b0418d3d2638e512","avatarUrl":"/avatars/55867c976c730935915311be8181ff32.svg","isPro":false,"fullname":"withinmiaov","user":"withinmiaov","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">

Papers

arxiv:2408.13467

LlamaDuo: LLMOps Pipeline for Seamless Migration from Service LLMs to Small-Scale Local LLMs

Published on Aug 24, 2024

· Submitted by

AK on Aug 27, 2024

Upvote

Authors:

Chansung Park ,

Juyong Jiang ,

Sayak Paul ,

Abstract

The LlamaDuo pipeline enables the transfer of knowledge from cloud-based large language models to smaller, locally manageable models through iterative fine-tuning with synthetic datasets.

AI-generated summary

The widespread adoption of cloud-based proprietary large language models (LLMs) has introduced significant challenges, including operational dependencies, privacy concerns, and the necessity of continuous internet connectivity. In this work, we introduce an LLMOps pipeline, "LlamaDuo", for the seamless migration of knowledge and abilities from service-oriented LLMs to smaller, locally manageable models. This pipeline is crucial for ensuring service continuity in the presence of operational failures, strict privacy policies, or offline requirements. Our LlamaDuo involves fine-tuning a small language model against the service LLM using a synthetic dataset generated by the latter. If the performance of the fine-tuned model falls short of expectations, it is enhanced by further fine-tuning with additional similar data created by the service LLM. This iterative process guarantees that the smaller model can eventually match or even surpass the service LLM's capabilities in specific downstream tasks, offering a practical and scalable solution for managing AI deployments in constrained environments. Extensive experiments with leading edge LLMs are conducted to demonstrate the effectiveness, adaptability, and affordability of LlamaDuo across various downstream tasks. Our pipeline implementation is available at https://github.com/deep-diver/llamaduo.