https://huggingface.co/TheFinAI/Fino1-8B

\n","updatedAt":"2025-02-13T02:45:28.975Z","author":{"_id":"63b58ed5889aa6707f0bb0f4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg","fullname":"Jimin Huang","name":"jiminHuang","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":28,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.49086400866508484},"editors":["jiminHuang"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg"],"reactions":[],"isReport":false}},{"id":"67ad66b2b51ac41873876bf1","author":{"_id":"63b58ed5889aa6707f0bb0f4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg","fullname":"Jimin Huang","name":"jiminHuang","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":28,"isUserFollowing":false},"createdAt":"2025-02-13T03:27:46.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"And also our dataset https://huggingface.co/datasets/TheFinAI/Fino1_Reasoning_Path_FinQA","html":"

And also our dataset https://huggingface.co/datasets/TheFinAI/Fino1_Reasoning_Path_FinQA

\n","updatedAt":"2025-02-13T03:27:46.859Z","author":{"_id":"63b58ed5889aa6707f0bb0f4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg","fullname":"Jimin Huang","name":"jiminHuang","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":28,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6693925857543945},"editors":["jiminHuang"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg"],"reactions":[],"isReport":false}},{"id":"67ae7eff926b659c7e8b4e01","author":{"_id":"64a02a724812ea6b2d3e18ac","avatarUrl":"/avatars/8c47baf0537872eacb93606d6d9d1bba.svg","fullname":"JM","name":"mukaj","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":10,"isUserFollowing":false},"createdAt":"2025-02-13T23:23:43.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"https://huggingface.co/mukaj/Llama-3.1-Hawkish-8B\n\nSimilar sized model also trained for financial reasoning, just tested it on FinQA and scored 60.94%, would be good to include in your leaderboard.","html":"

https://huggingface.co/mukaj/Llama-3.1-Hawkish-8B

Similar sized model also trained for financial reasoning, just tested it on FinQA and scored 60.94%, would be good to include in your leaderboard.

\n","updatedAt":"2025-02-13T23:23:43.182Z","author":{"_id":"64a02a724812ea6b2d3e18ac","avatarUrl":"/avatars/8c47baf0537872eacb93606d6d9d1bba.svg","fullname":"JM","name":"mukaj","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":10,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9541410803794861},"editors":["mukaj"],"editorAvatarUrls":["/avatars/8c47baf0537872eacb93606d6d9d1bba.svg"],"reactions":[],"isReport":false},"replies":[{"id":"67aecd5003b0a8a71d5f9376","author":{"_id":"65d50fd888f4a90589017398","avatarUrl":"/avatars/5e58259399ffe0d4e2e5a8c81fe65b30.svg","fullname":"Lingfei Qian","name":"lfqian","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false},"createdAt":"2025-02-14T04:57:52.000Z","type":"comment","data":{"edited":true,"hidden":false,"latest":{"raw":"Thank you for the comment. I have tested the performance of this model using FinQA dataset. But using our evaluation method from https://github.com/yale-nlp/DocMath-Eval that using GPT to extract results and compare the results, we only get 46.85%. May I ask what evaluation method are you using for this?","html":"

Thank you for the comment. I have tested the performance of this model using FinQA dataset. But using our evaluation method from https://github.com/yale-nlp/DocMath-Eval that using GPT to extract results and compare the results, we only get 46.85%. May I ask what evaluation method are you using for this?

\n","updatedAt":"2025-02-14T19:55:10.899Z","author":{"_id":"65d50fd888f4a90589017398","avatarUrl":"/avatars/5e58259399ffe0d4e2e5a8c81fe65b30.svg","fullname":"Lingfei Qian","name":"lfqian","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.9730063676834106},"editors":["lfqian"],"editorAvatarUrls":["/avatars/5e58259399ffe0d4e2e5a8c81fe65b30.svg"],"reactions":[],"isReport":false,"parentCommentId":"67ae7eff926b659c7e8b4e01"}}]},{"id":"67ae9d87fade8e6ee127686d","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":"2025-02-14T01:33:59.000Z","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* [Advancing Reasoning in Large Language Models: Promising Methods and Approaches](https://huggingface.co/papers/2502.03671) (2025)\n* [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://huggingface.co/papers/2501.12948) (2025)\n* [Enhancing Generalization in Chain of Thought Reasoning for Smaller Models](https://huggingface.co/papers/2501.09804) (2025)\n* [Baichuan4-Finance Technical Report](https://huggingface.co/papers/2412.15270) (2024)\n* [ChainRank-DPO: Chain Rank Direct Preference Optimization for LLM Rankers](https://huggingface.co/papers/2412.14405) (2024)\n* [FineMedLM-o1: Enhancing the Medical Reasoning Ability of LLM from Supervised Fine-Tuning to Test-Time Training](https://huggingface.co/papers/2501.09213) (2025)\n* [LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!](https://huggingface.co/papers/2502.07374) (2025)\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":"2025-02-14T01:33:59.774Z","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":0.7442139387130737},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2502.08127","authors":[{"_id":"67ad5ca29109885ce9b859e4","user":{"_id":"65d50fd888f4a90589017398","avatarUrl":"/avatars/5e58259399ffe0d4e2e5a8c81fe65b30.svg","isPro":false,"fullname":"Lingfei Qian","user":"lfqian","type":"user"},"name":"Lingfei Qian","status":"claimed_verified","statusLastChangedAt":"2025-06-11T20:21:33.656Z","hidden":false},{"_id":"67ad5ca29109885ce9b859e5","name":"Weipeng Zhou","hidden":false},{"_id":"67ad5ca29109885ce9b859e6","user":{"_id":"65d76cc5b9b7b8bf88faa916","avatarUrl":"/avatars/d95232cd0c307efab6197ade1a66190b.svg","isPro":true,"fullname":"Yan Wang","user":"YanAdjeNole","type":"user"},"name":"Yan Wang","status":"claimed_verified","statusLastChangedAt":"2025-05-28T19:15:18.741Z","hidden":false},{"_id":"67ad5ca29109885ce9b859e7","user":{"_id":"63a0c0803c8841cfe2cd1f15","avatarUrl":"/avatars/bbe216db7a33612f23d23ce4ed4ba3f9.svg","isPro":false,"fullname":"Xueqing Peng","user":"Xueqing","type":"user"},"name":"Xueqing Peng","status":"claimed_verified","statusLastChangedAt":"2025-06-03T09:14:22.042Z","hidden":false},{"_id":"67ad5ca29109885ce9b859e8","user":{"_id":"63b58ed5889aa6707f0bb0f4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg","isPro":true,"fullname":"Jimin Huang","user":"jiminHuang","type":"user"},"name":"Jimin Huang","status":"extracted_confirmed","statusLastChangedAt":"2025-02-19T03:28:40.861Z","hidden":false},{"_id":"67ad5ca29109885ce9b859e9","user":{"_id":"6479f4317c18dca75e9a9324","avatarUrl":"/avatars/9aa709230b057f57ee4415c04a622c63.svg","isPro":false,"fullname":"Xie","user":"QianqianXie1994","type":"user"},"name":"Qianqian Xie","status":"claimed_verified","statusLastChangedAt":"2025-02-13T08:22:01.539Z","hidden":false}],"publishedAt":"2025-02-12T05:13:04.000Z","submittedOnDailyAt":"2025-02-13T00:15:28.944Z","title":"Fino1: On the Transferability of Reasoning Enhanced LLMs to Finance","submittedOnDailyBy":{"_id":"63b58ed5889aa6707f0bb0f4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg","isPro":true,"fullname":"Jimin Huang","user":"jiminHuang","type":"user"},"summary":"Recent advancements in large language models (LLMs) have shown strong general\nreasoning abilities, yet their effectiveness in financial reasoning remains\nunderexplored. In this study, we comprehensively evaluate 16 powerful reasoning\nand general LLMs on three complex financial tasks involving financial text,\ntabular data, and equations, assessing numerical reasoning, tabular\ninterpretation, financial terminology comprehension, long-context processing,\nand equation-based problem solving. Our results show that while better datasets\nand pretraining improve financial reasoning, general enhancements like CoT\nfine-tuning do not always yield consistent gains. Moreover, all reasoning\nstrategies face challenges in improving performance on long-context and\nmulti-table tasks. To address these limitations, we develop a financial\nreasoning-enhanced model based on Llama-3.1-8B-Instruct, by CoT fine-tuning and\nreinforcement learning with domain-specific reasoning paths. Even with simple\nfine-tuning with one financial dataset, our model achieves a consistent 10%\nperformance improvement across tasks, surpassing all 8B models and even\nLlama3-70B-Instruct and Llama3.1-70B-Instruct on average. Our results highlight\nthe need for domain-specific adaptations in financial tasks, emphasizing future\ndirections such as multi-table reasoning, long-context processing, and\nfinancial terminology comprehension. All our datasets, models, and codes are\npublicly available. Furthermore, we introduce a leaderboard for benchmarking\nfuture datasets and models.","upvotes":59,"discussionId":"67ad5ca59109885ce9b85a5b","githubRepo":"https://github.com/the-finai/fino1","githubRepoAddedBy":"auto","ai_summary":"A study evaluates 16 large language models on complex financial tasks, finding that domain-specific CoT fine-tuning and reinforcement learning improve performance and highlight the need for further research on long-context and multi-table reasoning.","ai_keywords":["large language models","financial reasoning","CoT fine-tuning","reinforcement learning","domain-specific reasoning paths","multi-table reasoning","long-context processing"],"githubStars":69,"organization":{"_id":"658f4413674349122c0708e9","name":"TheFinAI","fullname":"The Fin AI","avatar":"https://cdn-uploads.huggingface.co/production/uploads/63b58ed5889aa6707f0bb0f4/ZK5nQKw34W3-eH3p4NAYc.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"63b58ed5889aa6707f0bb0f4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63b58ed5889aa6707f0bb0f4/znl74_aMswlV8VtHrfj3G.jpeg","isPro":true,"fullname":"Jimin Huang","user":"jiminHuang","type":"user"},{"_id":"63a0c0803c8841cfe2cd1f15","avatarUrl":"/avatars/bbe216db7a33612f23d23ce4ed4ba3f9.svg","isPro":false,"fullname":"Xueqing Peng","user":"Xueqing","type":"user"},{"_id":"64560c3f4a7ffb7d5a4686ab","avatarUrl":"/avatars/d0e625c3055fb19f22c82ffa17f95410.svg","isPro":false,"fullname":"Liu","user":"LuckyLiu","type":"user"},{"_id":"6479f4317c18dca75e9a9324","avatarUrl":"/avatars/9aa709230b057f57ee4415c04a622c63.svg","isPro":false,"fullname":"Xie","user":"QianqianXie1994","type":"user"},{"_id":"67ad638af634c181dcdb6cbf","avatarUrl":"/avatars/f5ceeb27794415a60d8076e6066ee5ad.svg","isPro":false,"fullname":"nanzhao","user":"nanzhao1234","type":"user"},{"_id":"65d76cc5b9b7b8bf88faa916","avatarUrl":"/avatars/d95232cd0c307efab6197ade1a66190b.svg","isPro":true,"fullname":"Yan Wang","user":"YanAdjeNole","type":"user"},{"_id":"67ad66991fd89c02d29e5d89","avatarUrl":"/avatars/e25d25d311dba96846d4013fd5825a8e.svg","isPro":false,"fullname":"yckongxiabin","user":"yckong","type":"user"},{"_id":"634cabd104491d9f7111eea3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/634cabd104491d9f7111eea3/JoqlugwfD1aGkd-wZTmP7.jpeg","isPro":true,"fullname":"Haohang Li","user":"Acatsama","type":"user"},{"_id":"64f757c6016d60f3199ef5e6","avatarUrl":"/avatars/2659ba698081265d0480b08161718013.svg","isPro":false,"fullname":"Yangyang Yu","user":"ShirleyY","type":"user"},{"_id":"66f612b934b8ac9ffa44f084","avatarUrl":"/avatars/6836c122e19c66c90f1673f28b30d7f0.svg","isPro":false,"fullname":"Tang","user":"tommysally","type":"user"},{"_id":"63082bb7bc0a2a5ee2253523","avatarUrl":"/avatars/6cf8d12d16d15db1070fbea89b5b3967.svg","isPro":false,"fullname":"Kuo-Hsin Tu","user":"dapumptu","type":"user"},{"_id":"66f1c27d0c26057904d21624","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66f1c27d0c26057904d21624/qwyJASNMilM6x49zqoUqV.webp","isPro":false,"fullname":"Triantafillos Papadopoulos","user":"tpapa","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":1,"organization":{"_id":"658f4413674349122c0708e9","name":"TheFinAI","fullname":"The Fin AI","avatar":"https://cdn-uploads.huggingface.co/production/uploads/63b58ed5889aa6707f0bb0f4/ZK5nQKw34W3-eH3p4NAYc.jpeg"}}">

Papers

arxiv:2502.08127

Fino1: On the Transferability of Reasoning Enhanced LLMs to Finance

Published on Feb 12, 2025

· Submitted by

Jimin Huang on Feb 13, 2025

#1 Paper of the day

The Fin AI

Upvote

Authors:

Lingfei Qian ,

Yan Wang ,

Xueqing Peng ,

Jimin Huang ,

Qianqian Xie

Abstract

A study evaluates 16 large language models on complex financial tasks, finding that domain-specific CoT fine-tuning and reinforcement learning improve performance and highlight the need for further research on long-context and multi-table reasoning.

AI-generated summary

Recent advancements in large language models (LLMs) have shown strong general reasoning abilities, yet their effectiveness in financial reasoning remains underexplored. In this study, we comprehensively evaluate 16 powerful reasoning and general LLMs on three complex financial tasks involving financial text, tabular data, and equations, assessing numerical reasoning, tabular interpretation, financial terminology comprehension, long-context processing, and equation-based problem solving. Our results show that while better datasets and pretraining improve financial reasoning, general enhancements like CoT fine-tuning do not always yield consistent gains. Moreover, all reasoning strategies face challenges in improving performance on long-context and multi-table tasks. To address these limitations, we develop a financial reasoning-enhanced model based on Llama-3.1-8B-Instruct, by CoT fine-tuning and reinforcement learning with domain-specific reasoning paths. Even with simple fine-tuning with one financial dataset, our model achieves a consistent 10% performance improvement across tasks, surpassing all 8B models and even Llama3-70B-Instruct and Llama3.1-70B-Instruct on average. Our results highlight the need for domain-specific adaptations in financial tasks, emphasizing future directions such as multi-table reasoning, long-context processing, and financial terminology comprehension. All our datasets, models, and codes are publicly available. Furthermore, we introduce a leaderboard for benchmarking future datasets and models.