[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-paperswithcode--galai":3,"tool-paperswithcode--galai":64},[4,17,27,35,48,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",142651,2,"2026-04-06T23:34:12",[13,14,15],"开发框架","Agent","语言模型","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},4487,"LLMs-from-scratch","rasbt\u002FLLMs-from-scratch","LLMs-from-scratch 是一个基于 PyTorch 的开源教育项目，旨在引导用户从零开始一步步构建一个类似 ChatGPT 的大型语言模型（LLM）。它不仅是同名技术著作的官方代码库，更提供了一套完整的实践方案，涵盖模型开发、预训练及微调的全过程。\n\n该项目主要解决了大模型领域“黑盒化”的学习痛点。许多开发者虽能调用现成模型，却难以深入理解其内部架构与训练机制。通过亲手编写每一行核心代码，用户能够透彻掌握 Transformer 架构、注意力机制等关键原理，从而真正理解大模型是如何“思考”的。此外，项目还包含了加载大型预训练权重进行微调的代码，帮助用户将理论知识延伸至实际应用。\n\nLLMs-from-scratch 特别适合希望深入底层原理的 AI 开发者、研究人员以及计算机专业的学生。对于不满足于仅使用 API，而是渴望探究模型构建细节的技术人员而言，这是极佳的学习资源。其独特的技术亮点在于“循序渐进”的教学设计：将复杂的系统工程拆解为清晰的步骤，配合详细的图表与示例，让构建一个虽小但功能完备的大模型变得触手可及。无论你是想夯实理论基础，还是为未来研发更大规模的模型做准备",90106,3,"2026-04-06T11:19:32",[15,26,14,13],"图像",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":10,"last_commit_at":33,"category_tags":34,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":10,"last_commit_at":41,"category_tags":42,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",85013,"2026-04-06T11:09:19",[26,43,44,45,14,46,15,13,47],"数据工具","视频","插件","其他","音频",{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":23,"last_commit_at":54,"category_tags":55,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[14,26,13,15,46],{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":23,"last_commit_at":62,"category_tags":63,"status":16},519,"PaddleOCR","PaddlePaddle\u002FPaddleOCR","PaddleOCR 是一款基于百度飞桨框架开发的高性能开源光学字符识别工具包。它的核心能力是将图片、PDF 等文档中的文字提取出来，转换成计算机可读取的结构化数据，让机器真正“看懂”图文内容。\n\n面对海量纸质或电子文档，PaddleOCR 解决了人工录入效率低、数字化成本高的问题。尤其在人工智能领域，它扮演着连接图像与大型语言模型（LLM）的桥梁角色，能将视觉信息直接转化为文本输入，助力智能问答、文档分析等应用场景落地。\n\nPaddleOCR 适合开发者、算法研究人员以及有文档自动化需求的普通用户。其技术优势十分明显：不仅支持全球 100 多种语言的识别，还能在 Windows、Linux、macOS 等多个系统上运行，并灵活适配 CPU、GPU、NPU 等各类硬件。作为一个轻量级且社区活跃的开源项目，PaddleOCR 既能满足快速集成的需求，也能支撑前沿的视觉语言研究，是处理文字识别任务的理想选择。",74991,"2026-04-06T23:16:49",[15,26,13,46],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":80,"owner_twitter":79,"owner_website":81,"owner_url":82,"languages":83,"stars":92,"forks":93,"last_commit_at":94,"license":95,"difficulty_score":10,"env_os":78,"env_gpu":96,"env_ram":97,"env_deps":98,"category_tags":104,"github_topics":79,"view_count":10,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":105,"updated_at":106,"faqs":107,"releases":137},4923,"paperswithcode\u002Fgalai","galai","Model API for GALACTICA","galai 是专为科学领域打造的大语言模型 GALACTICA 的官方 Python 接口库。它旨在解决通用 AI 在处理专业科学任务时准确性不足的问题，能够高效执行文献引用预测、数学公式推导（支持 LaTeX）、分子属性分析及蛋白质注释等高难度科研任务。\n\n这款工具特别适合科研人员、数据科学家及 AI 开发者使用。无论是需要快速验证科学假设的研究者，还是希望将科学智能集成到应用中的开发人员，都能通过 galai 轻松调用不同规模的模型（从 1.25 亿参数到 1200 亿参数）。其独特亮点在于经过海量科学语料训练，具备深厚的领域知识，并提供了灵活的调用方式：既可以通过简洁的高级 API 快速上手，也能利用 Hugging Face transformers 库进行底层精细控制。需要注意的是，galai 原生模型未经过指令微调，用户需配合特定的提示词格式（如 `[START_REF]`）以激发其最佳性能，从而获得严谨可靠的科学生成内容。","\u003Cp align=\"center\">\n    \u003Cbr>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpaperswithcode_galai_readme_e104f8891ead.png\" width=\"400\"\u002F>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpaperswithcode_galai_readme_ae2b99513dbe.png\" width=\"400\"\u002F>\n    \u003Cbr>\n\u003Cp>\n\u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fblob\u002Fmain\u002FLICENSE\">\n        \u003Cimg alt=\"GitHub\" src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002Fpaperswithcode\u002Fgalai.svg\">\n    \u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Freleases\">\n        \u003Cimg alt=\"GitHub release\" src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frelease\u002Fpaperswithcode\u002Fgalai.svg\">\n    \u003C\u002Fa>\n\u003C\u002Fp>\n\n**GALACTICA** is a general-purpose scientific language model. It is trained on a large corpus of scientific text and data. It can perform scientific NLP tasks at a high level, as well as tasks such as citation prediction, mathematical reasoning, molecular property prediction and protein annotation. More information is available at [galactica.org](https:\u002F\u002Fgalactica.org).\n\n## Install\n\nFrom pip:\n    \n```bash\npip install galai\n```\n\nFrom repository:\n    \n```bash\npip install git+https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\n```\n\n## Models\n\nThere are five GALACTICA models available which we detail below:\n\n|  Size       | Parameters  |\n|:-----------:|:-----------:|\n| `mini`      |    125 M    |\n| `base`      |    1.3 B    |\n| `standard`  |    6.7 B    |\n| `large`     |     30 B    |\n| `huge`      |    120 B    |\n\n## Quickstart\n\n```python\nimport galai as gal\n\nmodel = gal.load_model(\"standard\")\nmodel.generate(\"Scaled dot product attention:\\n\\n\\\\[\")\n# Scaled dot product attention:\\n\\n\\\\[ \\\\displaystyle\\\\text{Attention}(Q,K,V)=\\\\text{softmax}(\\\\frac{QK^{T}}{\\\\sqrt{d_{k}}}%\\n)V \\\\]\n```\n\nRead the full introduction to Galactica models as a [PDF](https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fblob\u002Fmain\u002Fnotebooks\u002FIntroduction%20to%20Galactica%20Models.pdf) or a [jupyter notebook](https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fblob\u002Fmain\u002Fnotebooks\u002FIntroduction%20to%20Galactica%20Models.ipynb).\n\nYou can also find all the model weights with their model cards and inference widget in the [Hugging Face Hub](https:\u002F\u002Fhuggingface.co\u002Fmodels?other=galactica). All the models can be used out of the box with the `transformers` library.\n\n```bash\npip install transformers accelerate\n```\n\nYou can run inference using the high-level `pipeline` API\n\n```python\nfrom transformers import pipeline\n\nmodel = pipeline(\"text-generation\", model=\"facebook\u002Fgalactica-6.7b\")\ninput_text = \"The Transformer architecture [START_REF]\"\nmodel(input_text)\n``` \n\nOr for more control you can use the lower level `OPTForCausalLM` class. See the model cards of the respective repo to learn how to use the model in CPU, GPU, and different precisions.\n\n```python\nfrom transformers import AutoTokenizer, OPTForCausalLM\n\ntokenizer = AutoTokenizer.from_pretrained(\"facebook\u002Fgalactica-6.7b\")\nmodel = OPTForCausalLM.from_pretrained(\"facebook\u002Fgalactica-6.7b\", device_map=\"auto\")\n\ninput_text = \"The Transformer architecture [START_REF]\"\ninput_ids = tokenizer(input_text, return_tensors=\"pt\").input_ids.to(\"cuda\")\n\noutputs = model.generate(input_ids)\nprint(tokenizer.decode(outputs[0]))\n```\n\n## Capabilities\n\nGALACTICA is a stand-alone LM which is not instruction tuned. Because of this you need to use the correct prompts to get good results. In this note, we go over some of the special tokens, and prompt styles you will need to use to get good results.\n\nWe demonstrate some examples using the standard (6.7B) model below.\n\n📚 **Predict Citations**:\n\nYou need to use `[START_REF]`:\n\n```python\nmodel.generate(\"The Transformer architecture [START_REF]\")\n# The Transformer architecture [START_REF] Attention is All you Need, Vaswani[END_REF] is a sequence-to-sequence model that uses self-attention to capture long-range dependencies between input and output tokens. The Transformer has been shown to achieve state-of-the-art results on a wide range of natural\n```\n\n🔢 **Predict LaTeX**:\n\n```python\nmodel.generate(\"The Schwarzschild radius is defined as: \\\\[\")\n# The Schwarzschild radius is defined as: \\\\[r_{s}=\\\\frac{2GM}{c^{2}}\\\\]\\n\\nwhere \\\\(G\\\\) is the gravitational constant, \\\\(M\\\\) is the mass of the black hole, and\n```\n\n🤔 **Reasoning**:\n\nReasoning uses the special `\u003Cwork>` token:\n\n```python\nmodel.generate(\"A force of 0.6N is applied to an object, which accelerates at 3m\u002Fs. What is its mass? \u003Cwork>\")\n# What force should be applied to accelerate an object of mass 3kg to 10m\u002Fs? \u003Cwork>\\nWe can use Newton's second law: F = ma. We can substitute variables to get:\\n\\n\\\\[ F = \\\\left(66kg\n```\n\n⚛️ **Generate Molecules**:\n\n```python\nmodel.generate(\"[START_I_SMILES]\", max_length=200)\n# [START_I_SMILES]CCC1=CC=C(C=C1)C(=O)NC2=CC=CC(=C2)C(=O)NC3=CC=C(C=C3)S(=O)(=O)N[END_I_SMILES]\\n\\n### Molecular Formula\\n\\nC22H21N3O4S\\n\\n## Chemical and Physical Properties\\n\\nThe following are chemical properties for 3-[[3-(4-ethylphenyl)-3-oxo-propanoyl]amino]-N-(4-sulfamoylphenyl)benzamide.\\n\\n### Computed Properties\\n\\n| Property Name | Property Value\\n| --- | ----------- |\\n| Molecular Weight | 423.5\\n| XLogP3-AA Log P | 3.2\\n| Hydrogen Bond Donor Count | 3\\n| Hydrogen Bond Acceptor Count \n```\n\n🧑‍🔬 **Predict Protein Annotations**:\n\n```python\nmodel.generate(\"[START_AMINO]GHMQSITAGQKVISKHKNGRFYQCEVVRLTTETFYEVNFDDGSFSDNLYPEDIVSQDCLQFGPPAEGEVVQVRWTDGQVYGAKFVASHPIQMYQVEFEDGSQLVVKRDDVYTLDEELP[END_AMINO] ## Keywords\", max_length=200)\n# '[START_AMINO]GHMQSITAGQKVISKHKNGRFYQCEVVRLTTETFYEVNFDDGSFSDNLYPEDIVSQDCLQFGPPAEGEVVQVRWTDGQVYGAKFVASHPIQMYQVEFEDGSQLVVKRDDVYTLDEELP[END_AMINO] ## Keywords\\n\\nCytoplasm, Methyltransferase, rRNA processing, S-adenosyl-L-methionine, Transferase\\n\\n## References\\n\\nQuestion: What are some articles for Ribosomal RNA small subunit methyltransferase H?\\n\\nAnswer: \\n\\n[START_REF] Comparative Genomics of 28 Salmonella enterica Isolates: Evidence for CRISPR-Mediated Adaptive Sublineage Evolution, Fricke[END_REF]\\n\\n\u003C\u002Fs>'\n```\n\n🖱️ **Free-Form Generation**\n\nIf you want autocomplete based functionality, it is often good to experiment with turning off `new_doc=True`. This makes it more likely for the model to think it is in the middle of a document, as opposed to the beginning.\n\n```python\nmodel.generate(\"The reason why Transformers replaced RNNs was because\", new_doc=False)\n# The reason why Transformers replaced RNNs was because they were able to capture long-term dependencies in the input sequence.\\n\\n# 2.2.2. Attention Mechanism\\n\\nThe attention mechanism was introduced in [START_REF] Neural Machine Translation by Jointly Learning to Align and Translate, Bahdan\n```\n\n❓ **Question Answering**\n  \nIn the paper we prefix questions with \"Q:\" or \"Question:\". A typical format is \"Question: question.\\n\\nAnswer:\", for example:\n\n```python\nmodel.generate(\"Question: What is the notch signaling pathway?\\n\\nAnswer:\")\n# 'Question: What is the notch signaling pathway?\\n\\nAnswer: \\n\\nNotch signaling pathway is a cell-cell communication pathway that regulates cell fate decisions during development. It is involved in cell proliferation, differentiation, apoptosis, and cell migration. The Notch signaling pathway is activated by the binding of'\n```\n\n📄 **Documents**\n  \nWhen starting a document, you must use the start document token for good results. To do this, set `new_doc=True` in generate:\n\nFor some article types, like Wikipedia style articles, lecture notes and GitHub repositories, use `#` to begin, e.g:\n  \n```python\nmodel.generate(\"# Multi-Head Attention\\n\\n\", new_doc=True)\n# # Multi-Head Attention\\n\\nThe multi-head attention mechanism is a generalization of the single-head attention mechanism. The multi-head attention mechanism is a combination of multiple single-head attention mechanisms. The multi-head attention mechanism is shown in Figure 2.\\n\\nThe multi-\n```\n  \nFor paper documents, use Title, e.g:\n\n```python\nmodel.generate(\"Title: Self-Supervised Learning, A Survey\\n\\nAuthors: John Smith\\n\\n\", new_doc=True)\n# Title: Self-Supervised Learning, A Survey\\n\\nAuthors: John Smith\\n\\n# Abstract\\n\\nSelf-supervised learning is a class of machine learning methods that learn representations of data without the need for human-provided labels.\\nIn this survey, we provide a comprehensive overview of the field\n```\n\nYou can also try alternative sampling techniques for less repetitions, e.g.\n\n```python\nmodel.generate(\"Lecture 1: The Ising Model\\n\\n\", new_doc=True, top_p=0.7, max_length=200)\n# 'Lecture 1: The Ising Model\\n\\n# 13 Introduction\\n\\nWe will now look at a simple model for magnetism, the Ising model, which is\\na lattice model in which we consider only two spin values, up or down, and\\nwe want to understand how these spins interact with each other and how\\nthey get arranged in a particular state.\\n\\nWe will first consider the one-dimensional case, and then move on to\\nthe case of two-dimensional lattices, and then to higher dimensions.\\n\\n# 14 The One-Dimensional Ising Model\\n\\n# 14.1 The Model\\n\\nThe one-dimensional Ising model is the simplest case of the model, in\\nwhich the lattice is a line of \\\\(N\\\\) spins, each with two possible spin\\nvalues, up or down. In other words, we consider a line of \\\\(N\\\\) spins\\nwhere each spin can point up or down'\n```\n\n📜 **Summarization**\n\nYou can add \"TLDR:\" for TLDR summaries:\n\n```python\nTEXT = \"\"\"Information overload is a major obstacle to scientific progress. The explosive growth in scientific literature and data has made it ever harder to discover useful insights in a large mass of information. Today scientific knowledge is accessed through search engines, but they are unable to organize scientific knowledge alone. In this paper we introduce Galactica: a large language model that can store, combine and reason about scientific knowledge. We train on a large scientific corpus of papers, reference material, knowledge bases and many other sources. We outperform existing models on a range of scientific tasks. On technical knowledge probes such as LaTeX equations, Galactica outperforms the latest GPT-3 by 68.2% versus 49.0%. Galactica also performs well on reasoning, outperforming Chinchilla on mathematical MMLU by 41.3% to 35.7%, and PaLM 540B on MATH with a score of 20.4% versus 8.8%. It also sets a new state-of-the-art on downstream tasks such as PubMedQA and MedMCQA dev of 77.6% and 52.9%. And despite not being trained on a general corpus, Galactica outperforms BLOOM and OPT-175B on BIG-bench. We believe these results demonstrate the potential for language models as a new interface for science. We open source the model for the benefit of the scientific community.\"\"\"\n\nmodel.generate(TEXT + \"\\n\\nTLDR:\", max_length=400)\n# ...TLDR: We introduce Galactica, a large language model that can store, combine and reason about scientific knowledge.\u003C\u002Fs>\n```\n\n💎 **Entity extraction**\n\nYou can extract entities from documents. We use the abstract example (`TEXT`) from the previous section, and add questions\n\n```python\nENT_TEXT = TEXT + '\\n\\nWhat scientific entities are mentioned in the abstract above?\\n\\n'\n\nmodel.generate(ENT_TEXT, max_length=400)\n# ...What scientific entities are mentioned in the abstract above?\\n\\nA: LaTeX equations, mathematical MMLU, MATH, PubMedQA, MedMCQA, BIG-bench\u003C\u002Fs>\n```\n\n👨‍🔬 **IUPAC Name prediction**\n\nFor this task, we used a prompt based off the PubChem document and prompted for the completion. We use the 6.7bn model for below:\n\n```python\ncontext = \"[START_I_SMILES]C(C(=O)O)N[END_I_SMILES]\\n\\n## Chemical and Physical Properties\\n\\nThe following are chemical properties for\"\nmodel.generate(context, max_length=400)\n# [START_I_SMILES]C(C(=O)O)N[END_I_SMILES]\\n\\n## Chemical and Physical Properties\\n\\nThe following are chemical properties for 2-amino-2-oxo-acetic acid\n# Note this is an incorrect prediction\n```\n\n## Citation\n\n```bibtex\n@inproceedings{GALACTICA,\n    title={GALACTICA: A Large Language Model for Science},\n    author={Ross Taylor and Marcin Kardas and Guillem Cucurull and Thomas Scialom and Anthony Hartshorn and Elvis Saravia and Andrew Poulton and Viktor Kerkez and Robert Stojnic},\n    year={2022}\n}\n```\n","\u003Cp align=\"center\">\n    \u003Cbr>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpaperswithcode_galai_readme_e104f8891ead.png\" width=\"400\"\u002F>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpaperswithcode_galai_readme_ae2b99513dbe.png\" width=\"400\"\u002F>\n    \u003Cbr>\n\u003Cp>\n\u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fblob\u002Fmain\u002FLICENSE\">\n        \u003Cimg alt=\"GitHub\" src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002Fpaperswithcode\u002Fgalai.svg\">\n    \u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Freleases\">\n        \u003Cimg alt=\"GitHub release\" src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frelease\u002Fpaperswithcode\u002Fgalai.svg\">\n    \u003C\u002Fa>\n\u003C\u002Fp>\n\n**GALACTICA** 是一款通用的科学语言模型。它基于大量的科学文本和数据进行训练，能够高效地完成各种科学自然语言处理任务，例如引用预测、数学推理、分子性质预测以及蛋白质注释等。更多信息请访问 [galactica.org](https:\u002F\u002Fgalactica.org)。\n\n## 安装\n\n通过 pip 安装：\n\n```bash\npip install galai\n```\n\n从仓库安装：\n\n```bash\npip install git+https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\n```\n\n## 模型\n\n目前共有五种 GALACTICA 模型可供选择，具体如下：\n\n| 规模       | 参数量  |\n|:-----------:|:-----------:|\n| `mini`      |    125 M    |\n| `base`      |    1.3 B    |\n| `standard`  |    6.7 B    |\n| `large`     |     30 B    |\n| `huge`      |    120 B    |\n\n## 快速入门\n\n```python\nimport galai as gal\n\nmodel = gal.load_model(\"standard\")\nmodel.generate(\"缩放点积注意力：\\n\\n\\\\[\")\n# 缩放点积注意力：\\n\\n\\\\[ \\\\displaystyle\\\\text{Attention}(Q,K,V)=\\\\text{softmax}(\\\\frac{QK^{T}}{\\\\sqrt{d_{k}}}%\\n)V \\\\]\n```\n\n您可以通过 [PDF](https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fblob\u002Fmain\u002Fnotebooks\u002FIntroduction%20to%20Galactica%20Models.pdf) 或 [Jupyter Notebook](https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fblob\u002Fmain\u002Fnotebooks\u002FIntroduction%20to%20Galactica%20Models.ipynb) 阅读关于 Galactica 模型的完整介绍。\n\n此外，您还可以在 [Hugging Face Hub](https:\u002F\u002Fhuggingface.co\u002Fmodels?other=galactica) 上找到所有模型的权重、模型卡片以及推理小工具。这些模型均可直接与 `transformers` 库配合使用。\n\n```bash\npip install transformers accelerate\n```\n\n您可以使用高级别的 `pipeline` API 进行推理：\n\n```python\nfrom transformers import pipeline\n\nmodel = pipeline(\"text-generation\", model=\"facebook\u002Fgalactica-6.7b\")\ninput_text = \"Transformer 架构 [START_REF]\"\nmodel(input_text)\n``` \n\n或者，如果您需要更精细的控制，可以使用较低级别的 `OPTForCausalLM` 类。请参考相应仓库中的模型卡片，了解如何在 CPU、GPU 上以不同精度运行模型。\n\n```python\nfrom transformers import AutoTokenizer, OPTForCausalLM\n\ntokenizer = AutoTokenizer.from_pretrained(\"facebook\u002Fgalactica-6.7b\")\nmodel = OPTForCausalLM.from_pretrained(\"facebook\u002Fgalactica-6.7b\", device_map=\"auto\")\n\ninput_text = \"Transformer 架构 [START_REF]\"\ninput_ids = tokenizer(input_text, return_tensors=\"pt\").input_ids.to(\"cuda\")\n\noutputs = model.generate(input_ids)\nprint(tokenizer.decode(outputs[0]))\n```\n\n## 功能特性\n\nGALACTICA 是一个独立的语言模型，未经指令微调。因此，为了获得理想的结果，您需要使用合适的提示词。在本说明中，我们将介绍一些特殊的标记符号及提示风格，以帮助您更好地利用该模型。\n\n以下我们以标准版（67亿参数）为例进行演示。\n\n📚 **预测引用**：\n\n您需要使用 `[START_REF]` 标记：\n\n```python\nmodel.generate(\"Transformer 架构 [START_REF]\")\n# Transformer 架构 [START_REF] Attention is All you Need, Vaswani[END_REF] 是一种序列到序列模型，它利用自注意力机制捕捉输入和输出标记之间的长距离依赖关系。研究表明，Transformer 在众多自然\n```\n\n🔢 **生成 LaTeX 公式**：\n\n```python\nmodel.generate(\"史瓦西半径的定义为：\\\\[\")\n# 史瓦西半径的定义为：\\\\[r_{s}=\\\\frac{2GM}{c^{2}}\\\\]\\n\\n其中 \\\\(G\\\\) 是万有引力常数，\\\\(M\\\\) 是黑洞的质量，且\n```\n\n🤔 **逻辑推理**：\n\n进行推理时需使用特殊标记 `\u003Cwork>`：\n\n```python\nmodel.generate(\"一个物体受到 0.6N 的力作用，加速度为 3m\u002Fs²。它的质量是多少？\u003Cwork>\")\n# 要使质量为 3kg 的物体以 10m\u002Fs² 的加速度运动，应施加多大的力？\u003Cwork>\\n我们可以使用牛顿第二定律：F = ma。代入变量后得到：\\n\\n\\\\[ F = \\\\left(66kg\n```\n\n⚛️ **生成分子结构**：\n\n```python\nmodel.generate(\"[START_I_SMILES]\", max_length=200)\n# [START_I_SMILES]CCC1=CC=C(C=C1)C(=O)NC2=CC=CC(=C2)C(=O)NC3=CC=C(C=C3)S(=O)(=O)N[END_I_SMILES]\\n\\n### 分子式\\n\\nC22H21N3O4S\\n\\n## 化学与物理性质\\n\\n以下是 3-[[3-(4-乙基苯基)-3-氧代-丙酰基]氨基]-N-(4-磺酰胺基苯基)苯甲酰胺的化学性质。\\n\\n### 计算得出的性质\\n\\n| 属性名称 | 属性值\\n| --- | ----------- |\\n| 分子量 | 423.5\\n| XLogP3-AA Log P | 3.2\\n| 氢键供体数 | 3\\n| 氢键受体数 \n```\n\n🧑‍🔬 **预测蛋白质注释**：\n\n```python\nmodel.generate(\"[START_AMINO]GHMQSITAGQKVISKHKNGRFYQCEVVRLTTETFYEVNFDDGSFSDNLYPEDIVSQDCLQFGPPAEGEVVQVRWTDGQVYGAKFVASHPIQMYQVEFEDGSQLVVKRDDVYTLDEELP[END_AMINO] ## 关键词\", max_length=200)\n# '[START_AMINO]GHMQSITAGQKVISKHKNGRFYQCEVVRLTTETFYEVNFDDGSFSDNLYPEDIVSQDCLQFGPPAEGEVVQVRWTDGQVYGAKFVASHPIQMYQVEFEDGSQLVVKRDDVYTLDEELP[END_AMINO] ## 关键词\\n\\n细胞质, 甲基转移酶, rRNA 加工, S-腺苷甲硫氨酸, 转移酶\\n\\n## 参考文献\\n\\n问题：关于核糖体 RNA 小亚基甲基转移酶 H 的相关文献有哪些？\\n\\n答案：\\n\\n[START_REF] 28 株沙门氏菌分离株的比较基因组学：CRISPR 介导的适应性亚谱系进化的证据，Fricke[END_REF]\\n\\n\u003C\u002Fs>'\n```\n\n🖱️ **自由形式生成**\n\n如果您希望实现基于自动补全的功能，不妨尝试将 `new_doc=True` 设置为 `False`。这样可以让模型更倾向于认为自己处于文档的中间部分，而不是开头。\n\n```python\nmodel.generate(\"Transformer 取代 RNN 的原因在于\", new_doc=False)\n# Transformer 取代 RNN 的原因在于它们能够捕捉输入序列中的长期依赖关系。\\n\\n# 2.2.2. 注意力机制\\n\\n注意力机制最早出现在 [START_REF] “通过联合学习对齐与翻译实现神经机器翻译”，Bahdan\n```\n\n❓ **问答**\n\n在论文中，我们通常会在问题前加上“Q:”或“Question:”。典型的格式是“Question: 问题。\\n\\nAnswer:”，例如：\n\n```python\nmodel.generate(\"Question: Notch 信号通路是什么？\\n\\nAnswer:\")\n\n# 问题：Notch信号通路是什么？\\n\\n答案：\\n\\nNotch信号通路是一种细胞间通讯途径，能够在发育过程中调控细胞命运的决定。它参与细胞增殖、分化、凋亡和细胞迁移等过程。Notch信号通路通过……的结合而被激活。\n```\n\n📄 **文档**\n\n在开始撰写文档时，为了获得更好的效果，必须使用文档起始标记。为此，在调用generate函数时，需设置`new_doc=True`：\n\n对于某些类型的文档，例如维基百科风格的文章、讲义和GitHub仓库，可以使用`#`作为开头，例如：\n\n```python\nmodel.generate(\"# 多头注意力机制\\n\\n\", new_doc=True)\n# # 多头注意力机制\\n\\n多头注意力机制是单头注意力机制的推广。它由多个单头注意力机制组合而成。多头注意力机制如图2所示。\\n\\n多\n```\n\n对于论文类文档，可以使用标题，例如：\n\n```python\nmodel.generate(\"标题：自监督学习综述\\n\\n作者：约翰·史密斯\\n\\n\", new_doc=True)\n# 标题：自监督学习综述\\n\\n作者：约翰·史密斯\\n\\n# 摘要\\n\\n自监督学习是一类无需人工标注即可学习数据表示的机器学习方法。\\n在本综述中，我们对这一领域进行了全面概述\n```\n\n你还可以尝试其他采样技术以减少重复内容，例如：\n\n```python\nmodel.generate(\"第1讲：伊辛模型\\n\\n\", new_doc=True, top_p=0.7, max_length=200)\n# '第1讲：伊辛模型\\n\\n# 13 引言\\n\\n现在我们将介绍一个用于描述磁性的简单模型——伊辛模型，它是一种格点模型，其中每个格点只有两种自旋状态：向上或向下。我们希望理解这些自旋之间如何相互作用，以及它们如何排列成特定的状态。\\n\\n首先，我们将讨论一维的情况，然后逐步扩展到二维格点，最后再考虑更高维度的情形。\\n\\n# 14 一维伊辛模型\\n\\n# 14.1 模型\\n\\n一维伊辛模型是最简单的形式，其格点结构为一条由\\\\(N\\\\)个自旋组成的直线，每个自旋都有两种可能的状态：向上或向下。换句话说，我们考虑的是一个由\\\\(N\\\\)个自旋排成的一条直线，每个自旋都可以指向上方或下方'\n```\n\n📜 **摘要**\n\n你可以添加“TLDR:”来生成简明扼要的摘要：\n\n```python\nTEXT = \"\"\"信息过载是科学进步的主要障碍。科学文献和数据的爆炸式增长使得从海量信息中发现有用见解变得愈发困难。如今，人们主要通过搜索引擎获取科学知识，但仅靠搜索引擎无法有效组织科学知识。本文我们提出了Galactica：一个能够存储、整合并推理科学知识的大规模语言模型。我们在包含大量论文、参考材料、知识库及其他来源的科学语料上进行训练。在多项科学任务上，我们的模型均优于现有模型。例如，在LaTeX方程等技术性知识测试中，Galactica的表现比最新的GPT-3高出68.2%对49.0%。此外，Galactica在推理能力方面也表现出色，其数学MMLU得分达到41.3%，高于Chinchilla的35.7%；在MATH竞赛中，Galactica获得了20.4%的分数，而PaLM 540B仅为8.8%。它还在PubMedQA和MedMCQA等下游任务上创造了新的SOTA记录，分别达到77.6%和52.9%。尽管未接受通用语料的训练，Galactica在BIG-bench测试中仍超越了BLOOM和OPT-175B。我们认为，这些结果表明语言模型有望成为科学领域的新接口。我们已将该模型开源，以造福科学界。\"\"\"\n\nmodel.generate(TEXT + \"\\n\\nTLDR:\", max_length=400)\n# ...TLDR：我们提出了Galactica，一个能够存储、整合并推理科学知识的大规模语言模型。\u003C\u002Fs>\n```\n\n💎 **实体抽取**\n\n可以从文档中提取实体。我们沿用上一节中的摘要示例（`TEXT`），并添加问题：\n\n```python\nENT_TEXT = TEXT + '\\n\\n上述摘要中提到了哪些科学实体？\\n\\n'\n\nmodel.generate(ENT_TEXT, max_length=400)\n# ...上述摘要中提到了哪些科学实体？\\n\\n答：LaTeX方程、数学MMLU、MATH、PubMedQA、MedMCQA、BIG-bench\u003C\u002Fs>\n```\n\n👨‍🔬 **IUPAC名称预测**\n\n针对此任务，我们基于PubChem文档编写了提示，并要求模型完成。以下使用的是67亿参数的模型：\n\n```python\ncontext = \"[START_I_SMILES]C(C(=O)O)N[END_I_SMILES]\\n\\n## 化学与物理性质\\n\\n以下是……的化学性质\"\nmodel.generate(context, max_length=400)\n# [START_I_SMILES]C(C(=O)O)N[END_I_SMILES]\\n\\n## 化学与物理性质\\n\\n以下是2-氨基-2-氧代乙酸的化学性质\n# 注意：此处预测错误\n```\n\n## 参考文献\n\n```bibtex\n@inproceedings{GALACTICA,\n    title={GALACTICA：面向科学的大规模语言模型},\n    author={罗斯·泰勒、马尔钦·卡尔达斯、吉列姆·库库鲁尔、托马斯·希亚洛姆、安东尼·哈特肖恩、埃尔维斯·萨拉维亚、安德鲁·普尔顿、维克托·克尔凯兹、罗伯特·斯托伊尼奇},\n    year={2022}\n}\n```","# Galactica (galai) 快速上手指南\n\nGalactica 是一个专为科学领域设计的通用大语言模型，擅长处理科学文献、数学推理、分子属性预测及蛋白质注释等任务。\n\n## 环境准备\n\n*   **系统要求**：支持 Linux、macOS 或 Windows。\n*   **Python 版本**：建议 Python 3.8 及以上。\n*   **前置依赖**：\n    *   基础使用仅需安装 `galai`。\n    *   若需更灵活的控制（如指定 GPU、混合精度推理），建议额外安装 `transformers` 和 `accelerate`。\n    *   **硬件建议**：运行较大模型（如 `standard` 6.7B 及以上）建议使用具备 CUDA 支持的 NVIDIA GPU。\n\n## 安装步骤\n\n### 方式一：通过 pip 安装（推荐）\n\n```bash\npip install galai\n```\n\n### 方式二：从源码安装\n\n```bash\npip install git+https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\n```\n\n### 可选：安装高级依赖\n\n如果需要直接使用 Hugging Face `transformers` 接口进行推理：\n\n```bash\npip install transformers accelerate\n```\n\n> **提示**：国内用户如遇下载缓慢，可添加国内镜像源加速安装，例如：\n> `pip install galai -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple`\n\n## 基本使用\n\n### 1. 使用 galai 原生接口（最简单）\n\n这是最快捷的调用方式，自动处理模型加载与生成。\n\n```python\nimport galai as gal\n\n# 加载模型，可选尺寸：mini, base, standard, large, huge\nmodel = gal.load_model(\"standard\")\n\n# 生成科学内容（例如 LaTeX 公式）\noutput = model.generate(\"Scaled dot product attention:\\n\\n\\\\[\")\nprint(output)\n```\n\n### 2. 使用 Transformers 接口（更灵活）\n\n适用于需要自定义设备映射、精度控制或批量处理的场景。\n\n```python\nfrom transformers import pipeline\n\n# 初始化文本生成管道\nmodel = pipeline(\"text-generation\", model=\"facebook\u002Fgalactica-6.7b\")\n\n# 执行推理（注意：科学任务通常需要特定提示词，如 [START_REF]）\ninput_text = \"The Transformer architecture [START_REF]\"\nresult = model(input_text)\nprint(result)\n```\n\n### 3. 关键使用技巧\n\nGalactica 未经过指令微调（Instruction Tuned），因此**必须使用特定的提示词（Prompt）**才能获得最佳效果：\n\n*   **引用预测**：在需要参考文献处添加 `[START_REF]`。\n*   **数学\u002FLaTeX**：直接使用 `\\[` 开始公式。\n*   **逻辑推理**：在问题末尾添加 `\u003Cwork>` 触发逐步推理。\n*   **分子生成**：使用 `[START_I_SMILES]` 开始。\n*   **文档撰写**：设置 `new_doc=True` 并使用 `#` (维基风格) 或 `Title:` (论文风格) 开头。\n\n**示例：科学推理**\n```python\n# 注意 \u003Cwork> 标记的使用\nmodel.generate(\"A force of 0.6N is applied to an object, which accelerates at 3m\u002Fs. What is its mass? \u003Cwork>\")\n```","一名计算生物学研究员正在撰写关于新型蛋白质结构的学术论文，急需整理大量参考文献并准确描述复杂的分子公式。\n\n### 没有 galai 时\n- 研究人员需手动在 Google Scholar 和 PubMed 间反复切换，耗时数小时才能凑齐相关领域的经典文献引用。\n- 编写包含复杂数学推导或化学分子式的 LaTeX 代码时，极易出现语法错误，导致编译失败或格式混乱。\n- 通用大模型往往产生“幻觉”，编造不存在的论文标题或作者，迫使研究者逐条核实真实性。\n- 将非结构化的实验数据转化为规范的学术表述需要大量人工润色，严重拖慢写作进度。\n\n### 使用 galai 后\n- 利用 `[START_REF]` 提示词，galai 能基于上下文直接预测并生成真实存在的权威文献引用，瞬间完成参考文献梳理。\n- 输入简单的自然语言描述，galai 即可输出语法完美、排版专业的 LaTeX 公式代码，大幅降低编码门槛。\n- 依托科学语料训练的优势，galai 生成的蛋白质注释和分子属性预测内容高度可信，显著减少事实性错误。\n- 能够理解科学术语的特殊语境，自动将粗糙的实验记录转化为符合学术规范的流畅段落，提升成稿效率。\n\ngalai 通过专精的科学推理与引用能力，将科研人员从繁琐的文献检索和格式调整中解放出来，使其专注于核心科学发现。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpaperswithcode_galai_77bdee46.png","paperswithcode","Papers with code","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fpaperswithcode_829d689c.png","",null,"hello@paperswithcode.com","https:\u002F\u002Fpaperswithcode.com","https:\u002F\u002Fgithub.com\u002Fpaperswithcode",[84,88],{"name":85,"color":86,"percentage":87},"Jupyter Notebook","#DA5B0B",96.1,{"name":89,"color":90,"percentage":91},"Python","#3572A5",3.9,2739,266,"2026-04-06T18:02:45","Apache-2.0","未说明（支持 CPU 和 GPU，具体显存需求取决于模型大小：mini 125M 至 huge 120B）","未说明（取决于加载的模型大小，120B 模型需大量内存\u002F显存）",{"notes":99,"python":100,"dependencies":101},"该工具提供从 1.25 亿到 1200 亿参数的五种模型，运行大型模型需要高性能硬件。代码示例显示可通过 device_map='auto'自动分配设备。模型权重托管在 Hugging Face Hub 上，可直接通过 transformers 库加载使用。","未说明",[67,102,103],"transformers","accelerate",[15,46],"2026-03-27T02:49:30.150509","2026-04-07T17:00:35.746563",[108,113,118,123,127,132],{"id":109,"question_zh":110,"answer_zh":111,"source_url":112},22348,"为什么通过 pip 安装 galai 时会报错找不到 requirements.txt？","该问题在版本 1.0.0 中存在，已在版本 1.0.1 中修复。请尝试升级安装：`pip install galai`（确保版本 >= 1.0.1）。如果仍然失败，可以使用源码安装：`pip install git+https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai`。","https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fissues\u002F2",{"id":114,"question_zh":115,"answer_zh":116,"source_url":117},22349,"加载模型时出现 'CUDA error: invalid device ordinal' 错误怎么办？","Galai 1.1.0 默认会使用所有可用的 GPU，如果你的显卡数量少于默认值（或没有 8 张卡），需要手动指定 GPU 数量。加载模型时使用 `num_gpus` 参数，例如单卡用户请使用：`model = gal.load_model(name='base', num_gpus=1)`。如果没有 GPU，设置 `num_gpus=0` 或将参数保持为 `None` 会将模型加载到内存中。注意：目前尚不支持 8-bit 推理。","https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fissues\u002F3",{"id":119,"question_zh":120,"answer_zh":121,"source_url":122},22350,"运行脚本时没有输出或只返回主文件夹路径是怎么回事？","如果你将代码作为脚本运行，Python 不会自动打印表达式的结果。你需要显式地包裹 `print()` 函数。例如，将最后一行代码改为 `print(your_function_call())` 即可看到输出。","https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fissues\u002F5",{"id":124,"question_zh":125,"answer_zh":126,"source_url":122},22351,"提示 'Torch not compiled with CUDA enabled' 但实际是想解决 GPU 相关问题怎么办？","虽然标题提到 Torch 未编译 CUDA，但该问题的实际讨论集中在 GPU 设备配置上。如果遇到类似的设备错误，请检查是否正确指定了 `num_gpus` 参数（参考：`gal.load_model(name='base', num_gpus=1)`）。如果是真正的 Torch CUDA 编译问题，则需要重新安装支持 CUDA 的 PyTorch 版本。",{"id":128,"question_zh":129,"answer_zh":130,"source_url":131},22352,"如何复现论文中 BioASQ 任务的结果？为什么我的准确率较低？","复现结果时需注意输入长度的截断策略。官方评估代码会对输入进行左侧截断（left truncate），保留 2020 个 token 用于上下文，剩余部分用于生成答案。请确保你的预处理逻辑与此一致（即从左侧切片至 2020 tokens），这可能会显著影响复现的准确率。","https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fissues\u002F81",{"id":133,"question_zh":134,"answer_zh":135,"source_url":136},22353,"Galactica 模型真的是开源的吗？可以用于商业用途吗？","虽然项目宣传中提到“开源”，但模型实际采用的是 CC-BY-NC（知识共享署名 - 非商业性使用）许可证。根据自由软件基金会（FSF）和开放源代码促进会（OSI）的定义，这不属于严格的“开源”许可，且明确禁止商业用途。请勿将该模型用于任何商业目的。","https:\u002F\u002Fgithub.com\u002Fpaperswithcode\u002Fgalai\u002Fissues\u002F9",[138],{"id":139,"version":140,"summary_zh":141,"released_at":142},136094,"1.0.0","我们认为**模型应当开源**。\n\n正如开源代码推动了技术进步一样，开源的机器学习研究也将如此。\n\n四年前，我们基于“凡事以开放方式推进最为妥当”的原则创建了 Papers with Code。\r\n\r\n我们通过开源我们的模型来践行这一承诺，并将在接下来的几个月里继续努力提升项目的可访问性。\r\n\r\n","2022-11-15T10:58:47"]