CoT

Search documents
10% KV Cache实现无损数学推理!这个开源方法解决推理大模型「记忆过载」难题
量子位· 2025-06-16 04:50
R-KV团队 投稿 量子位 | 公众号 QbitAI 推理大模型虽好,但一个简单的算数问题能推理整整三页,还都是重复的"废话",找不到重点…… 链式思考(Chain-of-Thought,CoT)让LLM解题思路清晰可见,却也让推理长度指数级膨胀。 以DeepSeek-R1-Llama-8B为例,一道AIME数学题就能写出 3.2万 个Token:模型权重15.5GB,KV缓存再吃 4.1GB ——显存瞬间见底。 现有KV压缩方法(SnapKV、StreamingLLM、H2O等)主要针对 长输入 设计,可一旦模型在输出端开始"碎碎念",相似句子之间互相打高 分注意力,反而让"按注意力删低分"策略失灵: 造成关键步骤被误删、重复内容却被保留、准确率断崖式下跌等问题。 而R-KV通过以下步骤,在模型解码时实时压缩KV缓存来处理冗余的键/值(KV)标记,仅保留重要且非冗余的标记: 让"长时间推理"不再是奢侈品。 项目详情可见文末链接。 R-KV三步走:冗余识别+重要性评估+动态淘汰 一种可以把大模型的"碎碎念"转化为可控记忆条目的高效压缩方法,出现了! R-KV开源登场: 显存↓90%、吞吐×6.6、准确率=10 ...
ACL 2025|为什么你设计的 Prompt 会成功?新理论揭示大模型 Prompt 设计的奥秘与效能
机器之心· 2025-06-16 04:04
本文共同一作是张翔和曹峻泰。张翔是英属哥伦比亚大学研究生,主要研究兴趣集中在大模型推理和 AI for Science;曹峻泰是英属哥伦比亚大学研究生, 主要研究兴趣集中在大模型推理和可解释性研究;本文通讯作者是来自纽约大学石溪分校的助理教授尤晨羽,以及来自 Meta Gen AI 的研究员丁渡鉴。 近年来,大型语言模型(LLM)在自然语言处理领域取得了革命性进展。然而,其底层的 Transformer 架构在处理复杂推理任务时仍有不足。尽管「思维 链」(CoT)提示技术提供了一条实用路径,但多数方法依赖通用指令,导致提示工程高度依赖反复试验,缺乏理论指导。 图 1 :Prompt 模板深刻影响着答案空间的配置和导航方式。左侧展示了不同的 Prompt(如 Auto-Prompt、RL-Prompt)如何在「Prompt 空间」中进行搜索,而右侧则展示了在特定 Prompt 指导下,如何在「答案空间」中进行搜索以得到解决方案(如 Tree-of-Thought、Graph-of-Thought)。 来自英属哥伦比亚大学、纽约大学石溪分校和浙江大学的研究团队深入剖析了 Prompt 如何在 LLM 的 CoT ...
Niagen Bioscience (CDXC) FY Conference Transcript
2025-06-09 19:15
Summary of Niagen Bioscience Conference Call Company Overview - Niagen Bioscience is focused on the development and commercialization of nicotinamide riboside (Niagen), a molecule that elevates NAD levels in cells, which is crucial for energy metabolism and cellular repair [5][7][8] - The company has over 90 patents protecting its core molecule and has built a strong scientific foundation with extensive research [8][14] Core Points and Arguments - **Scientific Research and Clinical Studies**: Niagen has conducted 40 peer-reviewed clinical studies demonstrating the safety and efficacy of its product, particularly in age-related conditions [7][18][24] - **Unique Positioning**: Unlike many dietary supplement companies, Niagen is rooted in scientific research and intellectual property, which differentiates it from competitors that primarily engage in marketing [11][14][17] - **Regulatory Challenges**: The company faces strict regulations that prevent it from making health claims typically associated with drugs, which influences its marketing strategy [23][24] - **Market Strategy**: Niagen relies on word-of-mouth and expert endorsements rather than direct consumer claims to build trust and credibility [24][25] Recent Developments - **Product Launches**: The company launched the Niagen Plus line, which includes IV products and at-home injectables, generating significant media attention and interest from celebrities and influencers [34][35][40] - **Clinical Trials**: A phase three clinical study on Parkinson's disease with 400 participants is nearing completion, which could significantly impact the company's valuation and credibility in the biotech space [18][54] - **Financial Health**: Niagen is cash flow positive, has no debt, and does not require additional capital for its operations, which is atypical for biotech companies [52][54] Potential Risks and Concerns - **Market Competition**: The rise of fraudulent companies claiming to offer NAD products poses a risk to the overall market perception and could impact Niagen's reputation [27][29] - **Consumer Education**: There is a need to educate consumers about the differences between Niagen and other NAD products, particularly regarding bioavailability and efficacy [27][37] Future Outlook - **Injectable Products**: The company plans to introduce a self-injection product within the next year, tapping into the growing trend of self-administered health solutions [49][50] - **Pharmaceutical Opportunities**: Niagen is exploring drug indications beyond dietary supplements, which could enhance its market position and attract sophisticated investors [54] Conclusion - Niagen Bioscience is well-positioned in the NAD market with a strong scientific foundation, innovative products, and a clear strategy to navigate regulatory challenges while maintaining financial stability. The company's focus on research and quality sets it apart from competitors, and its future product developments could significantly expand its market reach and impact.
奥特曼ChatGPT用法错了!最新研究:要求“直接回答”降低准确率,思维链提示作用也在下降
量子位· 2025-06-09 03:52
Core Viewpoint - The recent research from Wharton School and other institutions reveals that the "direct answer" prompt favored by Ultraman significantly reduces model accuracy [1][9]. Group 1: CoT Prompt Findings - Adding Chain of Thought (CoT) commands in prompts does not enhance reasoning models and increases time and computational costs [2][6]. - For reasoning models, the accuracy improvement from CoT is minimal, with o3-mini showing only a 4.1% increase, while time consumption rose by 80% [6][23]. - Non-reasoning models show mixed results with CoT prompts, necessitating careful consideration of benefits versus costs [7][12]. Group 2: Experimental Setup - The research utilized the GPQA Diamond dataset, which includes graduate-level expert reasoning questions, to test various reasoning and non-reasoning models under different conditions [5][9]. - Each model was tested in three experimental environments: forced reasoning, direct answer, and default [10][11]. Group 3: Performance Metrics - Four metrics were used to evaluate the models: overall results, 100% accuracy, 90% accuracy, and 51% accuracy [12][19]. - For non-reasoning models, CoT prompts improved average scores and the "51% correct" metric, with Gemini Flash 2.0 showing the most significant improvement [12][13]. - However, in the 100% and 90% accuracy metrics, the inclusion of CoT prompts led to declines in performance for some models [14][20]. Group 4: Conclusion on CoT Usage - The study indicates that while CoT can improve overall accuracy, it also increases answer instability [15][22]. - For models like o3-mini and o4-mini, the performance gain from using CoT prompts is minimal, and for Gemini 2.5 Flash, all metrics declined [20][21]. - Default settings of models are suggested to be effective for users, as many advanced models already incorporate reasoning processes internally [25].
Achieve Life Sciences (ACHV) 2025 Conference Transcript
2025-06-04 22:30
Achieve Life Sciences (ACHV) Conference Summary Company Overview - Achieve Life Sciences is focused on addressing nicotine dependence through its drug cytisinicline, which is positioned as a potential blockbuster treatment for smoking and vaping cessation [3][4][5] Industry Context - The U.S. faces a significant public health crisis with approximately 46 million individuals affected by nicotine dependence, including 29 million smokers and 17 million vapers [3][4] - The vaping population has increased by 6 million in the last 24 months, highlighting a growing concern [4] Key Points and Arguments 1. **Market Opportunity**: - The addressable market for nicotine dependence treatment is substantial, with an estimated annual market opportunity of $11 billion based on previous treatments like Chantix [33][35] - There are 15.4 million annual attempts to quit smoking, indicating a high demand for effective cessation tools [33] 2. **Product Development**: - Cytisinicline is expected to be the first new treatment for nicotine dependence in nearly 20 years, with an NDA filing planned for the end of June 2025 and anticipated FDA approval by mid-2026 [5][22] - The drug has shown a differentiated efficacy profile in clinical trials, with odds ratios of 5.3 and 5.8 in two phase three trials, indicating a significantly higher likelihood of quitting compared to existing treatments [19][20] 3. **Clinical Trials**: - Cytisinicline has undergone two large-scale phase three clinical trials involving over 1,600 patients, demonstrating robust efficacy and tolerability [5][25] - A phase two clinical trial for vaping cessation showed an odds ratio of 2.65, with plans for a phase three trial starting in early 2025 [27][23] 4. **Regulatory Pathway**: - The NDA submission is expected to be accepted within 74 days, with a clear regulatory pathway outlined by the FDA [22][43] 5. **Commercial Strategy**: - Achieve Life Sciences plans a targeted digital marketing strategy focusing on primary care physicians and patients, leveraging the lack of competition in the smoking cessation market since Chantix became generic [30][39] - The company aims to change the narrative around nicotine dependence from a lack of willpower to a medical condition, enhancing awareness and acceptance of cytisinicline [35][36] 6. **Safety and Tolerability**: - Cytisinicline has a favorable safety profile, with side effects significantly lower than those of existing treatments like Chantix, particularly in terms of nausea and vomiting [20][21] Additional Insights - The company has a strong intellectual property position and a proven leadership team, which is expected to drive shareholder value [43][44] - There is potential for future indications beyond nicotine dependence, including cannabis use, although the focus remains on the immediate market opportunity [50][52] Conclusion - Achieve Life Sciences is positioned to make a significant impact in the nicotine dependence treatment market with cytisinicline, addressing a critical public health issue while leveraging a well-defined commercial strategy and strong clinical data [44][54]
This Monster Dividend Growth Stock Is Up 50% So Far This Year
The Motley Fool· 2025-06-01 10:35
Core Viewpoint - Philip Morris International has achieved a 50% total return in 2025, significantly outperforming the S&P 500 index, which remains flat this year [1]. Group 1: Business Transformation - The company has successfully pivoted from traditional cigarettes to alternative nicotine products, recognizing the global decline in cigarette usage [4]. - Philip Morris holds a dominant position in the heat-not-burn category with its Iqos brand, capturing a 77% volume share in its operating markets [4]. - In the nicotine pouch segment, the company leads with its Zyn brand, exhibiting similar market share characteristics [4]. Group 2: Financial Performance - In the last quarter, 42% of the company's revenue and 44% of gross profit were derived from smoke-free products, indicating a significant shift in its revenue composition [5]. - Overall revenue has increased to $38 billion over the last 12 months, reflecting the successful transition to alternative nicotine products [5]. Group 3: Market Conditions - The depreciation of the U.S. dollar, which has fallen from around 110 to under 100, is expected to enhance revenue in U.S. dollar terms for Philip Morris, as it primarily operates outside the U.S. [6]. - The company is positioned to benefit from this currency trend, which has contributed to the stock price increase at the start of 2025 [6]. Group 4: Traditional Tobacco Outlook - Despite the decline in cigarette usage globally, traditional tobacco products are still expected to generate cash flow for the company, particularly outside of China and the U.S. [9]. - In the last quarter, gross profit from combustibles grew by 5.3% year over year, demonstrating the continued viability of traditional tobacco in international markets [9][10]. Group 5: Valuation and Future Prospects - The stock's forward price-to-earnings (P/E) ratio has increased to 24 from 14 a year ago, and the dividend yield has decreased to 3% from nearly 6% [13]. - This rising valuation suggests that the extraordinary 50% returns may not be sustainable, but the stock remains a viable investment due to its solid dividend yield and growth potential [14]. - The combination of Iqos and Zyn growth, along with pricing power in traditional cigarettes, positions the company for potential double-digit revenue and earnings growth in the coming years [14][15].
GPT-Kline:MCoT与技术分析
HTSC· 2025-05-31 10:25
Investment Rating - The report does not explicitly state an investment rating for the industry or the specific technology discussed. Core Insights - The research explores the application of Multimodal Chain of Thought (MCoT) in investment research, particularly in technical analysis using K-line charts, leading to the development of an automated platform called GPT-Kline [1][4][13]. - MCoT enhances the reasoning capabilities of large models by combining multimodal understanding with logical reasoning, allowing for more sophisticated analysis of complex tasks [2][21]. - The O3 model, launched by OpenAI, demonstrates impressive image reasoning capabilities, marking a significant step towards achieving general artificial intelligence (AGI) [2][37]. Summary by Sections Multimodal Reasoning - Multimodal collaboration is essential for large models to progress towards AGI, requiring them to be proficient in various modalities beyond just language [17]. - MCoT represents a significant advancement, enabling models to think based on images rather than merely perceiving them [21][31]. Application in Investment Research - The report highlights the potential of MCoT in technical analysis, particularly with K-line charts, which encapsulate vital trading information and patterns suitable for analysis [3][42]. - The O3 model's application in technical analysis shows its ability to process K-line images, perform necessary pre-processing, and generate analytical reports [3][43]. Development of GPT-Kline - GPT-Kline integrates MCoT with the capabilities of large models to create a specialized tool for K-line technical analysis, automating the entire analysis process from drawing to reporting [4][65]. - The platform features a user-friendly web interface designed for intuitive interaction, allowing users to engage with the analysis process effectively [4][83]. Model Comparison and Performance - The report compares various large models, including OpenAI's GPT-4o and Gemini-2.5 series, assessing their capabilities in K-line analysis and identifying Gemini-2.5 Flash as a strong performer [66][96]. - The analysis results indicate that while OpenAI's models tend to be conservative in their outputs, the Gemini models provide more comprehensive and accurate annotations [95][96].
Better High-Yield Dividend Stock: Altria or British American Tobacco?
The Motley Fool· 2025-05-30 07:14
Core Viewpoint - The tobacco industry is evolving into the nicotine industry, with Altria Group and British American Tobacco being key players, but British American Tobacco is currently better positioned for growth and market share in smokeless products [2][10][12]. Company Comparison - Altria and British American Tobacco both offer high dividend yields around 7% and have similar financial metrics, but their growth prospects differ significantly [2][5]. - Altria primarily operates in the U.S. with its Marlboro brand, while British American Tobacco has a global presence and competes mainly with Philip Morris International [4]. Financial Health - Both companies generate sufficient free cash flow to cover dividends and have significant stakes in other companies, with Altria's stake in Anheuser-Busch InBev valued at approximately $11 billion and British American Tobacco's stake in ITC Limited valued at around $16 billion [7]. Industry Adaptation - The decline in traditional cigarette use has prompted both companies to invest in smokeless nicotine products, with British American Tobacco leading in the electronic vape market with a 40% market share and 13.2% of total revenue from new product categories in 2024 [9][10]. - Altria has struggled with its investments in smokeless products, reporting only $300 million in sales from new categories in 2024, which is just 1.2% of its total revenue [11]. Market Dynamics - The U.S. government’s crackdown on illegal vape products benefits both companies, but British American Tobacco is expected to gain more due to its strong market share in vaping [14]. - Altria faces challenges in maintaining its market leadership in next-generation nicotine products, which could weaken its business as cigarette volumes decline [15].
Better Buy: Walmart vs. Target Stock
The Motley Fool· 2025-05-26 08:30
Core Insights - Walmart has significantly outperformed Target over the past three years, with Walmart's stock rising over 140% while Target's stock has declined nearly 40% [2] Company Comparison - Walmart operates 10,784 stores across 19 countries, making it a larger and more globally diversified retailer compared to Target, which has only 1,981 stores all located in the U.S. [4][6] - Walmart generates substantial revenue from international markets, including Mexico, Canada, and China, while Target's operations are solely focused on the U.S. market [5][6] - Walmart has a growing digital advertising business, Walmart Connect, and has expanded its advertising ecosystem through acquisitions, such as Vizio [5] - Target operates a smaller advertising business called Roundel and has previously exited international markets, indicating a more limited growth strategy [6] Business Strategies - Walmart's core strategy focuses on "everyday low prices" and a higher mix of essentials and groceries, while Target targets more affluent consumers with slightly pricier goods [7][8] - Both retailers have launched private label brands to enhance gross margins and are expanding online, curbside, and in-store pickup options to compete with Amazon [9] Financial Performance - From fiscal 2022 to fiscal 2025, Walmart's revenue had a compound annual growth rate (CAGR) of 6%, with earnings per share (EPS) growing at a CAGR of 14% [10] - Walmart's comparable-store sales in the U.S. showed consistent growth, with increases of 6.4%, 6.6%, 5.6%, and 4.5% over the respective fiscal years [11] - For fiscal 2026, Walmart expects net sales growth of 3% to 4% and adjusted EPS growth of 13% to 17% [12] - In contrast, Target's revenue from fiscal 2021 to fiscal 2024 had a CAGR of only 0.1%, with EPS declining at a negative CAGR of 14% [13] - Target's comparable-store sales peaked at 12.7% in fiscal 2021 but have since declined, with expectations for flat comps and only 1% net sales growth for fiscal 2025 [14] Market Challenges - Both companies faced boycotts related to their diversity, equity, and inclusivity initiatives, but Target's domestic focus may make it more vulnerable [15] - Target's smaller size and reliance on the U.S. market could hinder its ability to negotiate with overseas suppliers amid tariff pressures [16] Valuation Outlook - Walmart's stock is valued at 37 times forward earnings with a forward dividend yield of 1%, while Target's stock is cheaper at 11 times forward earnings and offers a higher yield of 4.7% [17] - Despite Target's lower valuation, Walmart is expected to continue outperforming due to its size, growth rate, diversification, and clearer business strategy [18]
22nd Century Group (XXII) Conference Transcript
2025-05-21 17:00
22nd Century Group (XXII) Conference May 21, 2025 12:00 PM ET Speaker0 Everyone next we have twenty second Century Group Inc. It trades on the NASDAQ under the symbol XXII and is the pioneering nicotine harm reduction company in the tobacco industry enabling smokers to take control of their nicotine consumption. Happy to welcome back Chairman and CEO, Larry Firestone. Welcome back to the conference today, Larry. Speaker1 Thanks Anna. Thanks for having us back and good morning everyone. So welcome to twenty ...