{"title":"InvestLM:一个使用金融领域指令调优的大型投资语言模型","authors":"Yi Yang, Yixuan Tang, Kar Yan Tam","doi":"arxiv-2309.13064","DOIUrl":null,"url":null,"abstract":"We present a new financial domain large language model, InvestLM, tuned on\nLLaMA-65B (Touvron et al., 2023), using a carefully curated instruction dataset\nrelated to financial investment. Inspired by less-is-more-for-alignment (Zhou\net al., 2023), we manually curate a small yet diverse instruction dataset,\ncovering a wide range of financial related topics, from Chartered Financial\nAnalyst (CFA) exam questions to SEC filings to Stackexchange quantitative\nfinance discussions. InvestLM shows strong capabilities in understanding\nfinancial text and provides helpful responses to investment related questions.\nFinancial experts, including hedge fund managers and research analysts, rate\nInvestLM's response as comparable to those of state-of-the-art commercial\nmodels (GPT-3.5, GPT-4 and Claude-2). Zero-shot evaluation on a set of\nfinancial NLP benchmarks demonstrates strong generalizability. From a research\nperspective, this work suggests that a high-quality domain specific LLM can be\ntuned using a small set of carefully curated instructions on a well-trained\nfoundation model, which is consistent with the Superficial Alignment Hypothesis\n(Zhou et al., 2023). From a practical perspective, this work develops a\nstate-of-the-art financial domain LLM with superior capability in understanding\nfinancial texts and providing helpful investment advice, potentially enhancing\nthe work efficiency of financial professionals. We release the model parameters\nto the research community.","PeriodicalId":501372,"journal":{"name":"arXiv - QuantFin - General Finance","volume":"65 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2023-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"InvestLM: A Large Language Model for Investment using Financial Domain Instruction Tuning\",\"authors\":\"Yi Yang, Yixuan Tang, Kar Yan Tam\",\"doi\":\"arxiv-2309.13064\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"We present a new financial domain large language model, InvestLM, tuned on\\nLLaMA-65B (Touvron et al., 2023), using a carefully curated instruction dataset\\nrelated to financial investment. Inspired by less-is-more-for-alignment (Zhou\\net al., 2023), we manually curate a small yet diverse instruction dataset,\\ncovering a wide range of financial related topics, from Chartered Financial\\nAnalyst (CFA) exam questions to SEC filings to Stackexchange quantitative\\nfinance discussions. InvestLM shows strong capabilities in understanding\\nfinancial text and provides helpful responses to investment related questions.\\nFinancial experts, including hedge fund managers and research analysts, rate\\nInvestLM's response as comparable to those of state-of-the-art commercial\\nmodels (GPT-3.5, GPT-4 and Claude-2). Zero-shot evaluation on a set of\\nfinancial NLP benchmarks demonstrates strong generalizability. From a research\\nperspective, this work suggests that a high-quality domain specific LLM can be\\ntuned using a small set of carefully curated instructions on a well-trained\\nfoundation model, which is consistent with the Superficial Alignment Hypothesis\\n(Zhou et al., 2023). From a practical perspective, this work develops a\\nstate-of-the-art financial domain LLM with superior capability in understanding\\nfinancial texts and providing helpful investment advice, potentially enhancing\\nthe work efficiency of financial professionals. We release the model parameters\\nto the research community.\",\"PeriodicalId\":501372,\"journal\":{\"name\":\"arXiv - QuantFin - General Finance\",\"volume\":\"65 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2023-09-15\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - QuantFin - General Finance\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2309.13064\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - QuantFin - General Finance","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2309.13064","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
摘要
我们提出了一个新的金融领域大型语言模型,InvestLM,在llama - 65b上进行了调整(Touvron等人,2023),使用了精心策划的与金融投资相关的指令数据集。受“少即是多”(Zhouet al., 2023)的启发,我们手动策划了一个小而多样的指令数据集,涵盖了广泛的金融相关主题,从特许金融分析师(CFA)考试问题到SEC文件,再到Stackexchange定量金融讨论。InvestLM在理解金融文本方面表现出很强的能力,并对投资相关问题提供了有益的回答。包括对冲基金经理和研究分析师在内的金融专家认为,ateinvestlm的反应与最先进的商业模型(GPT-3.5、GPT-4和Claude-2)相当。在一组金融NLP基准上的零射击评估显示出很强的泛化性。从研究的角度来看,这项工作表明,一个高质量的特定领域的法学硕士可以在一个训练有素的基础模型上使用一组精心策划的指令,这与表面对齐假设是一致的(Zhou et al., 2023)。从实际应用的角度来看,本研究开发的金融领域法学硕士具有卓越的金融文本理解能力和提供有用的投资建议,有可能提高金融专业人士的工作效率。我们向研究界发布模型参数。
InvestLM: A Large Language Model for Investment using Financial Domain Instruction Tuning
We present a new financial domain large language model, InvestLM, tuned on
LLaMA-65B (Touvron et al., 2023), using a carefully curated instruction dataset
related to financial investment. Inspired by less-is-more-for-alignment (Zhou
et al., 2023), we manually curate a small yet diverse instruction dataset,
covering a wide range of financial related topics, from Chartered Financial
Analyst (CFA) exam questions to SEC filings to Stackexchange quantitative
finance discussions. InvestLM shows strong capabilities in understanding
financial text and provides helpful responses to investment related questions.
Financial experts, including hedge fund managers and research analysts, rate
InvestLM's response as comparable to those of state-of-the-art commercial
models (GPT-3.5, GPT-4 and Claude-2). Zero-shot evaluation on a set of
financial NLP benchmarks demonstrates strong generalizability. From a research
perspective, this work suggests that a high-quality domain specific LLM can be
tuned using a small set of carefully curated instructions on a well-trained
foundation model, which is consistent with the Superficial Alignment Hypothesis
(Zhou et al., 2023). From a practical perspective, this work develops a
state-of-the-art financial domain LLM with superior capability in understanding
financial texts and providing helpful investment advice, potentially enhancing
the work efficiency of financial professionals. We release the model parameters
to the research community.