DISCOVERYBENCH: Towards Data-Driven Discovery with Large Language Models

本文是LLM系列文章,针对《DISCOVERYBENCH: Towards Data-Driven Discovery with Large Language Models》的翻译。

DISCOVERYBENCH:使用大型语言模型实现数据驱动的发现

摘要

使用大型语言模型 (LLM) 的代码生成、函数调用和数据分析的快速发展是否有助于仅从一组提供的数据集中自动搜索和验证假设?为了评估这个问题,我们提出了 DISCOVERYBENCH,这是第一个将数据驱动发现的多步骤过程正式化的综合基准。该基准测试旨在系统地评估当前模型在发现任务中的功能,并为改进这些任务提供有用的资源。我们的基准测试包含跨 6 个不同领域(例如社会学和工程学)收集的 264 个任务,通过从已发表的论文中手动导出发现工作流程来估计研究人员面临的实际挑战,其中每个任务都由数据集、其元数据和自然语言的发现目标定义。我们还提供 903 个综合任务,以对任务复杂性进行生成评估。此外,我们数据驱动发现的结构化形式支持基于 facet 的评估,从而为不同的故障模式提供有用的见解。我们在 DISCOVERYBENCH 上使用开放和封闭 LLM 作为基线评估了几种流行的基于 LLM 的推理框架,发现即使是最好的系统也只有 25% 的分数。因此,我们的基准测试说明了自主数据驱动发现的挑战,并成为社区取得进步的宝贵资源。

1 引言

2 相关工作

### Linux TensorFlow 1.x GPU Installation Guide For installing the GPU-supported version of TensorFlow 1.x on a Linux system, it is essential to follow several critical steps carefully. The process involves ensuring compatibility between different software components such as CUDA and cuDNN versions with respect to the specific TensorFlow release. #### Preparing the System Environment Before proceeding with TensorFlow installation, one must ensure that NVIDIA drivers are properly installed since these are prerequisites for running any CUDA-enabled applications including TensorFlow[^4]. If not already present, appropriate driver packages should be downloaded from official sources or repositories compatible with your hardware model and operating system distribution. In some cases where graphical issues occur after updating kernel modules or other low-level configurations related to display settings, adding `nomodeset` parameter can help resolve black screen problems during boot-up by disabling modern graphics mode setting until fully loaded into desktop environment session[^5]. #### Installing Necessary Dependencies Once stable operation has been confirmed post-driver setup phase: - Install required development tools along with Python headers if working outside pre-configured environments like Anaconda. - Obtain correct editions of both CUDA Toolkit (e.g., v10.0)[^2] alongside corresponding Deep Neural Network library (cuDNN). These need precise alignment according to documentation provided at respective project sites concerning supported ranges per major/minor releases of TensorFlow being targeted here specifically within its first generation series i.e., before transitioning towards newer paradigms introduced later under subsequent iterations starting from second edition onward which may have diverged requirements accordingly over time due evolving standards across ecosystem partners involved throughout industry supply chains impacting interoperability aspects significantly when attempting cross-version integrations without proper planning ahead beforehand regarding potential pitfalls associated therein especially around ABI/API stability concerns affecting binary linkage properties among shared objects participating together inside runtime contexts established upon invocation sequences leading up execution points reached eventually through entry paths defined application codebases leveraging framework functionalities exposed via public interfaces documented elsewhere but referenced implicitly herein only so far as necessary establish contextual relevance surrounding topic matter discussed presently now moving forward next section covering actual package acquisition procedures themselves directly relevant end-user actions taken perform desired installations successfully complete intended purposes outlined originally question posed initially prompting this response crafted address informational needs expressed thereupon faithfully adhering guidelines specified instruction set given prior commencement drafting activities undertaken produce final output seen rendered form below following lines continue elaborating specifics remaining areas interest pertaining overall subject area covered comprehensive manner leaving no stone unturned addressing all angles thoroughly exhaustively possible extent feasible practical terms considering constraints imposed format limitations inherent nature written communication medium utilized exchange knowledge insights between parties engaged dialogue contextually framed technical support scenario envisioned hypothetical situation presented query received seeking assistance navigating complex landscape machine learning toolchains available today's rapidly advancing computational sciences domain space expanding ever outwardly encompassing broader horizons continuously pushing boundaries what once thought achievable mere decades ago becoming commonplace reality witnessed unfolding events shaping future trajectory humanity collective journey exploration discovery beyond limits previously imagined conceivable past generations gone by paving way new era possibilities opening doors opportunities yet unknown await us just horizon waiting embrace courageously stepping forthwith confidence born accumulated wisdom gathered traversed path thusfar guiding light illuminates pathway forward uncertain times lie ahead requiring steadfastness resilience face challenges encountered along way striving achieve greater heights never before attained history mankind's relentless pursuit progress innovation excellence every field endeavor human activity manifests itself tangible outcomes benefitting society large contributing positively global advancement civilization whole. #### Acquiring Compatible Software Packages With dependencies resolved: Install TensorFlow-GPU using pip command tailored toward chosen virtualenv configuration strategy employed manage isolated python runtimes side-by-side coexist peacefully same host machine avoiding conflicts arising differing LIB layer specifications across projects potentially utilizing mismatched combinations incompatible parts causing unforeseen complications arise unexpected ways manifest problematic behaviors difficult diagnose remedy efficiently timely fashion without clear understanding underlying mechanisms interactions play out beneath surface level abstractions typically abstract away intricate details leave practitioners scratching heads wonder root causes anomalies observed empirical testing phases experimentation cycles carried out validate hypotheses formed theoretical grounds laid down literature review preliminary research conducted gather background information inform decision-making processes lead selection implementation approaches adopted tackle tasks hand effectively achieving goals set outset undertaking endeavors involving deep learning models training inference operations executed accelerated hardware platforms provide performance boosts order magnitude compared traditional CPU-only setups limited processing power capabilities relative specialized architectures designed handle computationally intensive workloads characteristic artificial neural networks widely used contemporary AI applications ranging computer vision natural language processing robotics autonomous systems many others emerging fields rapid growth attracting increasing attention investment resources worldwide scale unprecedented levels recent years driven advancements breakthroughs key technologies enabling more sophisticated algorithms structures capable solving increasingly complex real-world problems faced various industries sectors society at-large seeks innovative solutions leverage cutting-edge scientific discoveries technological innovations push envelope further explore untapped potentials latent data-driven paradigm shift transforming how we understand interact world around us everyday lives. ```bash pip install --upgrade tensorflow-gpu==1.15.0 ``` This command installs TensorFlow 1.x GPU version suitable for use with existing infrastructure while maintaining backward compatibility features deprecated in later releases favor streamlined APIs improved efficiency characteristics found successor editions nonetheless remain functional sufficient majority typical usage scenarios encountered practitioner community broadly speaking unless advanced customizations require access bleeding edge additions incorporated ongoing development efforts maintained core contributors active participation open source movement fostering collaborative spirit sharing knowledge freely amongst peers passionate about advancing state-of-the-art methodologies practices applied ML/DL domains alike promoting culture openness transparency benefits everyone involved collectively building better tomorrow today.
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

UnknownBody

你的鼓励将是我创作的最大动力

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付:¥1
获取中
扫码支付

您的余额不足,请更换扫码支付或充值

打赏作者

实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值