Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456 Paper page - MUSCLE: A Model Update Strategy for Compatible LLM Evolution
\n","updatedAt":"2024-07-15T02:38:51.502Z","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":0.49586430191993713},"editors":["akhaliq"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg"],"reactions":[],"isReport":false}},{"id":"669606b572c02268e2f5a196","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":"2024-07-16T05:35:49.000Z","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* [RE-Adapt: Reverse Engineered Adaptation of Large Language Models](https://huggingface.co/papers/2405.15007) (2024)\n* [Low-Redundant Optimization for Large Language Model Alignment](https://huggingface.co/papers/2406.12606) (2024)\n* [On the Transformations across Reward Model, Parameter Update, and In-Context Prompt](https://huggingface.co/papers/2406.16377) (2024)\n* [Increasing Model Capacity for Free: A Simple Strategy for Parameter Efficient Fine-tuning](https://huggingface.co/papers/2407.01320) (2024)\n* [MeteoRA: Multiple-tasks Embedded LoRA for Large Language Models](https://huggingface.co/papers/2405.13053) (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.
\n
The following papers were recommended by the Semantic Scholar API
Please give a thumbs up to this comment if you found it helpful!
\n
If you want recommendations for any Paper on Hugging Face checkout this Space
\n
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend
\n","updatedAt":"2024-07-16T05:35:49.690Z","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":0.7339421510696411},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2407.09435","authors":[{"_id":"66948ba1df5d51613344d461","name":"Jessica Echterhoff","hidden":false},{"_id":"66948ba1df5d51613344d462","name":"Fartash Faghri","hidden":false},{"_id":"66948ba1df5d51613344d463","name":"Raviteja Vemulapalli","hidden":false},{"_id":"66948ba1df5d51613344d464","name":"Ting-Yao Hu","hidden":false},{"_id":"66948ba1df5d51613344d465","name":"Chun-Liang Li","hidden":false},{"_id":"66948ba1df5d51613344d466","name":"Oncel Tuzel","hidden":false},{"_id":"66948ba1df5d51613344d467","user":{"_id":"653a8e65c75dc136bfb5b0f8","avatarUrl":"/avatars/ef252794236ce0fe2debf12773c95bb2.svg","isPro":false,"fullname":"Hadi Pouransari","user":"hpouransari","type":"user"},"name":"Hadi Pouransari","status":"extracted_pending","statusLastChangedAt":"2024-07-15T02:38:26.354Z","hidden":false}],"publishedAt":"2024-07-12T17:12:48.000Z","submittedOnDailyAt":"2024-07-15T01:08:51.476Z","title":"MUSCLE: A Model Update Strategy for Compatible LLM Evolution","submittedOnDailyBy":{"_id":"60f1abe7544c2adfd699860c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg","isPro":false,"fullname":"AK","user":"akhaliq","type":"user"},"summary":"Large Language Models (LLMs) are frequently updated due to data or\narchitecture changes to improve their performance. When updating models,\ndevelopers often focus on increasing overall performance metrics with less\nemphasis on being compatible with previous model versions. However, users often\nbuild a mental model of the functionality and capabilities of a particular\nmachine learning model they are interacting with. They have to adapt their\nmental model with every update -- a draining task that can lead to user\ndissatisfaction. In practice, fine-tuned downstream task adapters rely on\npretrained LLM base models. When these base models are updated, these\nuser-facing downstream task models experience instance regression or negative\nflips -- previously correct instances are now predicted incorrectly. This\nhappens even when the downstream task training procedures remain identical. Our\nwork aims to provide seamless model updates to a user in two ways. First, we\nprovide evaluation metrics for a notion of compatibility to prior model\nversions, specifically for generative tasks but also applicable for\ndiscriminative tasks. We observe regression and inconsistencies between\ndifferent model versions on a diverse set of tasks and model updates. Second,\nwe propose a training strategy to minimize the number of inconsistencies in\nmodel updates, involving training of a compatibility model that can enhance\ntask fine-tuned language models. We reduce negative flips -- instances where a\nprior model version was correct, but a new model incorrect -- by up to 40% from\nLlama 1 to Llama 2.","upvotes":23,"discussionId":"66948ba2df5d51613344d4c2","ai_summary":"The work provides evaluation metrics and a training strategy to minimize inconsistencies and negative flips in Large Language Model updates, ensuring better compatibility with prior model versions for downstream tasks.","ai_keywords":["Large Language Models (LLMs)","generative tasks","discriminative tasks","instance regression","negative flips","compatibility model","task fine-tuned language models"]},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"664d21b07e7b513c7e695413","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/664d21b07e7b513c7e695413/4xIs59LsYa5hg5JPEO9-D.jpeg","isPro":false,"fullname":"Sudhir","user":"sudzdpn","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"646fce0528638f11a83ee890","avatarUrl":"/avatars/6bbe81608f9fb82506dec7cbd182d94b.svg","isPro":false,"fullname":"Hristo Panev","user":"hppdqdq","type":"user"},{"_id":"64e567c9ddbefb63095a9662","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/F2BwrOU0XpzVI5nd-TL54.png","isPro":false,"fullname":"Bullard ","user":"Charletta1","type":"user"},{"_id":"62ced8629b96f22525b9cdf5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62ced8629b96f22525b9cdf5/g23JBCoMKBgsFB8fb0pw9.jpeg","isPro":false,"fullname":"YYY","user":"zzfive","type":"user"},{"_id":"667045d0f108d832f01bb8aa","avatarUrl":"/avatars/8b892ede91d002ee25755546f12efad2.svg","isPro":false,"fullname":"Yedidia AGNIMO","user":"Yedson54","type":"user"},{"_id":"648eb1eb59c4e5c87dc116e0","avatarUrl":"/avatars/c636cea39c2c0937f01398c94ead5dad.svg","isPro":false,"fullname":"fdsqefsgergd","user":"T-representer","type":"user"},{"_id":"65e0ca2a045ac5cccc3956c3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/i2NY9Gb3tCEC6EQXpJ766.png","isPro":false,"fullname":"Raul Rodriguez Abad","user":"r3abad","type":"user"},{"_id":"651c80a26ba9ab9b9582c273","avatarUrl":"/avatars/e963452eafd21f517d800f2e58e0f918.svg","isPro":false,"fullname":"siyeng feng","user":"siyengfeng","type":"user"},{"_id":"6374ff24cc5cc31768847b8c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6374ff24cc5cc31768847b8c/7_hKYVzSE226qzB5wJ6gw.jpeg","isPro":false,"fullname":"Minghui Jia","user":"Maxwell-Jia","type":"user"},{"_id":"6695da1fc73fcf7a04ea4d8d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/9s5pXvXU9Edk-ubmdHwNV.png","isPro":false,"fullname":"KomiyaJapan","user":"KomiyaJapan","type":"user"},{"_id":"640ae2aeb871d7117d5f8135","avatarUrl":"/avatars/af0820f65b8db3180075bbdcb15624e0.svg","isPro":false,"fullname":"Trangle Heshvp","user":"Trangle","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">
The work provides evaluation metrics and a training strategy to minimize inconsistencies and negative flips in Large Language Model updates, ensuring better compatibility with prior model versions for downstream tasks.
AI-generated summary
Large Language Models (LLMs) are frequently updated due to data or
architecture changes to improve their performance. When updating models,
developers often focus on increasing overall performance metrics with less
emphasis on being compatible with previous model versions. However, users often
build a mental model of the functionality and capabilities of a particular
machine learning model they are interacting with. They have to adapt their
mental model with every update -- a draining task that can lead to user
dissatisfaction. In practice, fine-tuned downstream task adapters rely on
pretrained LLM base models. When these base models are updated, these
user-facing downstream task models experience instance regression or negative
flips -- previously correct instances are now predicted incorrectly. This
happens even when the downstream task training procedures remain identical. Our
work aims to provide seamless model updates to a user in two ways. First, we
provide evaluation metrics for a notion of compatibility to prior model
versions, specifically for generative tasks but also applicable for
discriminative tasks. We observe regression and inconsistencies between
different model versions on a diverse set of tasks and model updates. Second,
we propose a training strategy to minimize the number of inconsistencies in
model updates, involving training of a compatibility model that can enhance
task fine-tuned language models. We reduce negative flips -- instances where a
prior model version was correct, but a new model incorrect -- by up to 40% from
Llama 1 to Llama 2.