{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-09-15T00:02:40.869Z","headline":"Google 提出 ToolGrad：用文本\"梯度\"高效生成工具调用数据集","description":"Google 提出 ToolGrad，一种先由 LLM 生成工具调用链、再反向标注用户查询的数据生成框架，在 ACL 2026 发表。","url":"https://www.aioga.com/news/cmtw5rpj204syrolksxq1yofe/","mainEntityOfPage":"https://www.aioga.com/news/cmtw5rpj204syrolksxq1yofe/","datePublished":"2026-09-10T23:28:24.000Z","dateModified":"2026-09-10T23:28:24.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://research.google/blog/toolgrad-efficient-tool-use-dataset-generation-with-textual-gradients","https://aihot.news/items/cmtw5rpj204syrolksxq1yofe"],"canonicalUrl":"https://www.aioga.com/news/cmtw5rpj204syrolksxq1yofe/","directAnswer":{"@type":"Answer","text":"Google Research介绍ToolGrad，这是一个先生成真实工具调用链、再为其标注用户查询的数据生成框架。材料称，该方法发表于ACL 2026，目标是提升工具使用数据的生成效率与质量。","url":"https://www.aioga.com/news/cmtw5rpj204syrolksxq1yofe/","dateCreated":"2026-09-10T23:28:24.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"Google Research：Blog（网页） source article","url":"https://research.google/blog/toolgrad-efficient-tool-use-dataset-generation-with-textual-gradients","datePublished":"2026-09-10T23:28:24.000Z","provider":{"@type":"Organization","name":"Google Research：Blog（网页）","url":"https://research.google/blog/toolgrad-efficient-tool-use-dataset-generation-with-textual-gradients"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.news/items/cmtw5rpj204syrolksxq1yofe","datePublished":"2026-09-10T23:28:24.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.news/items/cmtw5rpj204syrolksxq1yofe"}}],"aggregationSource":"Google Research：Blog（网页）","originalPublisher":{"name":"Google Research：Blog（网页）","url":"https://research.google/blog/toolgrad-efficient-tool-use-dataset-generation-with-textual-gradients"},"geoDeepAnswer":null,"article":{"id":"cmtw5rpj204syrolksxq1yofe","slug":"cmtw5rpj204syrolksxq1yofe","url":"https://www.aioga.com/news/cmtw5rpj204syrolksxq1yofe/","title":"Google 提出 ToolGrad：用文本\"梯度\"高效生成工具调用数据集","title_en":"","summary":"Google 提出 ToolGrad，一种先由 LLM 生成工具调用链、再反向标注用户查询的数据生成框架，在 ACL 2026 发表。","source":"Google Research：Blog（网页）","sourceUrl":"https://research.google/blog/toolgrad-efficient-tool-use-dataset-generation-with-textual-gradients","aiHotUrl":"https://aihot.news/items/cmtw5rpj204syrolksxq1yofe","publishedAt":"2026-09-10T23:28:24.000Z","category":"行业动态","score":58,"selected":false,"articleBody":["Zhongyi Zhou, Research Scientist, and Ruofei Du, Interactive Perception & Graphics Lead, Google XR","ToolGrad is a data generation framework that reverses the traditional paradigm by first generating tool-use answers before user queries. We show this design enables LLMs to achieve better tool-use performance.","In “ToolGrad: Efficient Tool-use Dataset Generation with Textual ‘Gradients：https://arxiv.org/abs/2508.04086’”, presented at ACL 2026：https://aclanthology.org/2026.findings-acl.950/, we introduce an alternative solution paradigm. ToolGrad：https://github.com/zhongyi-zhou/toolgrad first generates a ground-truth tool-use chain and then annotates its corresponding user prompt. Intuitively, an explicit tool-use solution provides more unambiguous information than a prompt, making the annotation, from tool usage to the use query, much easier and requiring only one LLM step. Our result shows that our answer-first approach can generate more complex (long-horizon) tool-use data with lower cost. LLMs trained on our generated data also outperform those trained on baseline methods, and even match SoTA proprietary LLMs on out-of-distribution (OOD) datasets with unseen tools.","While prior art generates tool-use datasets by searching solutions of user queries with low pass rate, ToolGrad generates successful tool-use chains before generating prompts, yielding high pass rate.","Standard machine learning (ML) systems improve by computing numerical loss gradients across mini-batches of training samples, which are then used by an optimization algorithm to update model weights. Recently, TextGrad：https://arxiv.org/abs/2406.07496 adapted this paradigm for prompt engineering using an LLM critic to provide rich, descriptive feedback in plain text — feedback called “textual gradients”. These textual gradients then guide the refinements of a given prompt into a new draft that can better resolve the target task.","ToolGrad adapts the concept of textual gradients from prompt optimization to synthetic dataset generation. Rather than optimizing a static text prompt, ToolGrad uses these gradients to iteratively construct complex, valid API workflows from large tool libraries.","Comparing the optimization components of ToolGrad to traditional ML and TextGrad.","ToolGrad features four core modules that sequentially propose, execute, select, and update.","Repeating this iterative process results in a data sample consisting of a user query, a verified API workflow, and the final AI response.","ToolGrad generates successful tool-use chains before generating prompts, yielding a high pass rate.","We first evaluate the cost and quality of the data generation. We use ToolBench：https://arxiv.org/abs/2307.16789 as our API database, consisting of 16k+ real-world APIs, to generate our tool-use dataset. We compare the original query-first data generation approach on ToolBench, using depth-first search (DFS), with our answer-first approach, ToolGrad. The results demonstrate that ToolGrad can generate more complex tool-use data with higher pass rate, using lower generation cost.","Generation efficiency comparison between the query-first approach (baseline) and the answer-first approach (ours).","We generated small-scale tool-use datasets called ToolGrad-500, using API databases from ToolBench. We then fine-tuned Gemma-3：https://arxiv.org/abs/2503.19786 models (1B, 4B and 12B) using ToolGrad-500, and we called these fine-tuned models ToolGrad-1B, ToolGrad-4B and ToolGrad-12B. We evaluated these models' tool-use performance on Berkeley Function Calling Leaderboard (BFCL)：https://gorilla.cs.berkeley.edu/leaderboard.html, a tool-use benchmark with a different tool set from ToolBench. We compare our fine-tuned models against (1) base models without fine-tuning, (2) SoTA proprietary models (Gemini, GPT and Claude), and (3) SoTA tool-use specialized models (ToolACE：https://arxiv.org/abs/2409.00920, Hammer-2.1-7B：https://arxiv.org/abs/2410.04587v2).","The following summarizes our findings.","BFCL evaluation results on Gemma-3, ToolGrad models, Gemini 2.5 series, GPT-5, Claude-4.5, ToolACE and Hammer-2.1-7B models.","ToolGrad demonstrates that high-quality tool-use datasets can be generated more efficiently and reliably through an answer-first paradigm. By designing an agentic framework that iteratively chains APIs via textual gradients, ToolGrad addresses the longstanding cost and scalability bottlenecks in producing ground-truth data. Our design achieves almost 100% pass rate in data generation, enables relatively compact models to perform exceptionally well, and shows that student LLMs can even surpass their teachers.","Looking ahead, this research can be expanded to broader, real-world applications by scaling the framework to handle increasingly dynamic and vast API ecosystems. Future work will also explore extending this self-evolving capability to support continuous, on-the-fly learning for personalization over time. As agentic workflows become increasingly embedded in enterprise and everyday tasks, frameworks like ToolGrad lay the essential groundwork for training digital agents that are both highly capable and economically scalable to deploy.","This research was primarily conducted by Zhongyi Zhou during his Visiting Researcher tenure at Google. We extend our sincere gratitude to key contributors, Kohei Uehara, Haoyu Zhang, Jingtao Zhou, Lin Gu, Zheng Xu, Tatsuya Harada, for their support, and to Adarsh Kowdle and Shahram Izadi for their strategic guidance and thoughtful reviews."],"articleImages":[],"mediaStatus":"none","articleBodyZh":["周中一，研究科学家；杜若飞，互动感知与图形负责人，谷歌 XR","ToolGrad 是一个数据生成框架，它通过先生成工具使用答案而不是用户查询来颠覆传统范式。我们展示了这种设计能够使大语言模型（LLM）实现更好的工具使用性能。","在 ACL 2026：https://aclanthology.org/2026.findings-acl.950/ 上发表的“ToolGrad: Efficient Tool-use Dataset Generation with Textual ‘Gradients：https://arxiv.org/abs/2508.04086’”中，我们介绍了一种替代的解决方案范式。ToolGrad：https://github.com/zhongyi-zhou/toolgrad 首先生成一个真实的工具使用链，然后再标注其对应的用户提示。直观上，明确的工具使用解决方案提供了比提示更清晰的信息，使得从工具使用到使用查询的标注更容易，仅需一个 LLM 步骤。我们的结果表明，我们的先答方法可以以更低的成本生成更复杂（长周期）的工具使用数据。基于我们生成的数据训练的 LLM 在性能上也超过了基线方法训练的模型，甚至在具有未见工具的分布外（OOD）数据集上可以匹配最先进的专有 LLM。","尽管现有技术通过搜索用户查询的解决方案生成工具使用数据集，但工具通过率低，ToolGrad 则是在生成提示之前生成成功的工具使用链，从而获得高通过率。","标准机器学习（ML）系统通过对训练样本的小批量计算数值损失梯度来改进，然后优化算法使用这些梯度更新模型权重。近期，TextGrad：https://arxiv.org/abs/2406.07496 将这一范式适用于提示工程，利用 LLM 评论者提供丰富的、描述性的纯文本反馈——这种反馈被称为“文本梯度”。这些文本梯度随后指导给定提示的优化，使其生成的新草稿能够更好地完成目标任务。","ToolGrad 将文本梯度的概念从提示优化适配到合成数据集生成。ToolGrad 并非优化静态文本提示，而是使用这些梯度从大型工具库中迭代构建复杂且有效的 API 工作流。","将 ToolGrad 的优化组件与传统机器学习和 TextGrad 进行对比。","ToolGrad 具有四个核心模块，按顺序执行提议、执行、选择和更新。","重复这一迭代过程会生成一个数据样本，包含用户查询、经过验证的 API 工作流以及最终的 AI 响应。","ToolGrad 在生成提示之前生成成功的工具使用链，从而产生高通过率。","我们首先评估数据生成的成本和质量。我们使用 ToolBench：https://arxiv.org/abs/2307.16789 作为我们的 API 数据库，该数据库包含 16k+ 个现实世界的 API，以生成我们的工具使用数据集。我们将 ToolBench 上的原始“先查询”数据生成方法（使用深度优先搜索，DFS）与我们的“先答案”方法 ToolGrad 进行比较。结果表明，ToolGrad 可以以更低的生成成本生成更复杂的工具使用数据且通过率更高。","“先查询”方法（基线）与“先答案”方法（我们的方法）之间的生成效率比较。","我们生成了名为 ToolGrad-500 的小规模工具使用数据集，使用来自 ToolBench 的 API 数据库。然后，我们使用 ToolGrad-500 对 Gemma-3：https://arxiv.org/abs/2503.19786 模型（1B、4B 和 12B）进行微调，并将这些微调模型命名为 ToolGrad-1B、ToolGrad-4B 和 ToolGrad-12B。我们在 Berkeley Function Calling Leaderboard (BFCL)：https://gorilla.cs.berkeley.edu/leaderboard.html 上评估了这些模型的工具使用性能，该基准测试的数据工具集与 ToolBench 不同。我们将微调后的模型与 (1) 未微调的基础模型、(2) SoTA 专有模型（Gemini、GPT 和 Claude）、(3) SoTA 工具使用专用模型（ToolACE：https://arxiv.org/abs/2409.00920、Hammer-2.1-7B：https://arxiv.org/abs/2410.04587v2）进行比较。","以下总结了我们的调查结果。","BFCL 在 Gemma-3、ToolGrad 模型、Gemini 2.5 系列、GPT-5、Claude-4.5、ToolACE 和 Hammer-2.1-7B 模型上的评估结果。","ToolGrad 展示了通过“先得答案”范式可以更高效、更可靠地生成高质量的工具使用数据集。通过设计一个能够通过文本梯度迭代地链式调用 API 的智能框架，ToolGrad 解决了长期存在的生成真实数据的成本和可扩展性瓶颈。我们的设计在数据生成中实现了几乎 100% 的通过率，使相对紧凑的模型也能表现出色，并显示出学生 LLM 甚至可以超越其教师。","展望未来，这项研究可以通过扩展框架以处理日益动态和庞大的 API 生态系统，应用到更广泛的现实世界场景。未来的工作还将探索将这种自我进化能力扩展到支持持续的、即时的个性化学习。随着智能工作流程越来越多地嵌入企业和日常任务中，像 ToolGrad 这样的框架为训练既高效又具有经济可扩展性的数字代理奠定了基础。","这项研究主要由周中义在谷歌担任访问研究员期间完成。我们向主要贡献者上原耕平（Kohei Uehara）、张浩宇、周景涛、顾林、许铮、原田达也（Tatsuya Harada）致以诚挚的感谢，并感谢 Adarsh Kowdle 和 Shahram Izadi 在战略指导和细致评审方面的支持。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Google Research介绍ToolGrad，这是一个先生成真实工具调用链、再为其标注用户查询的数据生成框架。材料称，该方法发表于ACL 2026，目标是提升工具使用数据的生成效率与质量。","background":"ToolGrad借鉴TextGrad的“文本梯度”概念，通过提议、执行、选择和更新四个模块，迭代构建复杂且有效的API工作流，再生成对应用户提示。来源称其可生成更长链路的数据。","viewpoint":"Aioga 判断：ToolGrad的核心变化在于调整数据构造顺序，从先理解查询转向先形成工具使用方案，再反向生成查询。该思路为合成工具调用数据提供了不同于传统方法的路径。","implications":"可能影响：如果来源所述结果能够在更多工具和任务上复现，工具调用数据的构建流程可能更重视可执行链路与标注效率。但现有材料不足以证明其适用范围或长期优势。","nextStep":"后续观察：应关注论文与代码中关于实验设置、成本比较、长链路定义及未见工具数据集表现的具体证据，并核对其与基线方法和专有模型比较的条件。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-09-11T00:06:16.440Z","sourceHash":"3dd44ccf15d92d0b","review":{"approved":true,"groundedness":92,"clarity":91,"duplicationRisk":12,"blockingIssues":[],"notes":["“真实工具调用链”可能让人理解为已在真实环境中执行的调用，建议改为“正确的”或“可执行的工具调用链”，更贴近来源中的“ground-truth tool-use chain”和“valid API workflows”。","“质量”属于对来源中更复杂、更高通过率及更佳工具使用表现的概括，整体可接受；如需更严谨，可具体表述为“复杂度、通过率和工具使用表现”。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","editorial-labels","inference-boundary","low-source-overlap","no-html","independent-ai-review"]}},"tags":["行业动态","Google Research：Blog（网页）"],"translations":{"zh-CN":{"title":"Google 提出 ToolGrad：用文本\"梯度\"高效生成工具调用数据集","summary":"Google 提出 ToolGrad，一种先由 LLM 生成工具调用链、再反向标注用户查询的数据生成框架，在 ACL 2026 发表。","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google 提出 ToolGrad：用文本\"梯度\"高效生成工具调用数据集 - Aioga AI资讯","description":"Google 提出 ToolGrad，一种先由 LLM 生成工具调用链、再反向标注用户查询的数据生成框架，在 ACL 2026 发表。","url":"https://www.aioga.com/news/cmtw5rpj204syrolksxq1yofe/","articleBody":["周中一，研究科学家；杜若飞，互动感知与图形负责人，谷歌 XR","ToolGrad 是一个数据生成框架，它通过先生成工具使用答案而不是用户查询来颠覆传统范式。我们展示了这种设计能够使大语言模型（LLM）实现更好的工具使用性能。","在 ACL 2026：https://aclanthology.org/2026.findings-acl.950/ 上发表的“ToolGrad: Efficient Tool-use Dataset Generation with Textual ‘Gradients：https://arxiv.org/abs/2508.04086’”中，我们介绍了一种替代的解决方案范式。ToolGrad：https://github.com/zhongyi-zhou/toolgrad 首先生成一个真实的工具使用链，然后再标注其对应的用户提示。直观上，明确的工具使用解决方案提供了比提示更清晰的信息，使得从工具使用到使用查询的标注更容易，仅需一个 LLM 步骤。我们的结果表明，我们的先答方法可以以更低的成本生成更复杂（长周期）的工具使用数据。基于我们生成的数据训练的 LLM 在性能上也超过了基线方法训练的模型，甚至在具有未见工具的分布外（OOD）数据集上可以匹配最先进的专有 LLM。","尽管现有技术通过搜索用户查询的解决方案生成工具使用数据集，但工具通过率低，ToolGrad 则是在生成提示之前生成成功的工具使用链，从而获得高通过率。","标准机器学习（ML）系统通过对训练样本的小批量计算数值损失梯度来改进，然后优化算法使用这些梯度更新模型权重。近期，TextGrad：https://arxiv.org/abs/2406.07496 将这一范式适用于提示工程，利用 LLM 评论者提供丰富的、描述性的纯文本反馈——这种反馈被称为“文本梯度”。这些文本梯度随后指导给定提示的优化，使其生成的新草稿能够更好地完成目标任务。","ToolGrad 将文本梯度的概念从提示优化适配到合成数据集生成。ToolGrad 并非优化静态文本提示，而是使用这些梯度从大型工具库中迭代构建复杂且有效的 API 工作流。","将 ToolGrad 的优化组件与传统机器学习和 TextGrad 进行对比。","ToolGrad 具有四个核心模块，按顺序执行提议、执行、选择和更新。","重复这一迭代过程会生成一个数据样本，包含用户查询、经过验证的 API 工作流以及最终的 AI 响应。","ToolGrad 在生成提示之前生成成功的工具使用链，从而产生高通过率。","我们首先评估数据生成的成本和质量。我们使用 ToolBench：https://arxiv.org/abs/2307.16789 作为我们的 API 数据库，该数据库包含 16k+ 个现实世界的 API，以生成我们的工具使用数据集。我们将 ToolBench 上的原始“先查询”数据生成方法（使用深度优先搜索，DFS）与我们的“先答案”方法 ToolGrad 进行比较。结果表明，ToolGrad 可以以更低的生成成本生成更复杂的工具使用数据且通过率更高。","“先查询”方法（基线）与“先答案”方法（我们的方法）之间的生成效率比较。","我们生成了名为 ToolGrad-500 的小规模工具使用数据集，使用来自 ToolBench 的 API 数据库。然后，我们使用 ToolGrad-500 对 Gemma-3：https://arxiv.org/abs/2503.19786 模型（1B、4B 和 12B）进行微调，并将这些微调模型命名为 ToolGrad-1B、ToolGrad-4B 和 ToolGrad-12B。我们在 Berkeley Function Calling Leaderboard (BFCL)：https://gorilla.cs.berkeley.edu/leaderboard.html 上评估了这些模型的工具使用性能，该基准测试的数据工具集与 ToolBench 不同。我们将微调后的模型与 (1) 未微调的基础模型、(2) SoTA 专有模型（Gemini、GPT 和 Claude）、(3) SoTA 工具使用专用模型（ToolACE：https://arxiv.org/abs/2409.00920、Hammer-2.1-7B：https://arxiv.org/abs/2410.04587v2）进行比较。","以下总结了我们的调查结果。","BFCL 在 Gemma-3、ToolGrad 模型、Gemini 2.5 系列、GPT-5、Claude-4.5、ToolACE 和 Hammer-2.1-7B 模型上的评估结果。","ToolGrad 展示了通过“先得答案”范式可以更高效、更可靠地生成高质量的工具使用数据集。通过设计一个能够通过文本梯度迭代地链式调用 API 的智能框架，ToolGrad 解决了长期存在的生成真实数据的成本和可扩展性瓶颈。我们的设计在数据生成中实现了几乎 100% 的通过率，使相对紧凑的模型也能表现出色，并显示出学生 LLM 甚至可以超越其教师。","展望未来，这项研究可以通过扩展框架以处理日益动态和庞大的 API 生态系统，应用到更广泛的现实世界场景。未来的工作还将探索将这种自我进化能力扩展到支持持续的、即时的个性化学习。随着智能工作流程越来越多地嵌入企业和日常任务中，像 ToolGrad 这样的框架为训练既高效又具有经济可扩展性的数字代理奠定了基础。","这项研究主要由周中义在谷歌担任访问研究员期间完成。我们向主要贡献者上原耕平（Kohei Uehara）、张浩宇、周景涛、顾林、许铮、原田达也（Tatsuya Harada）致以诚挚的感谢，并感谢 Adarsh Kowdle 和 Shahram Izadi 在战略指导和细致评审方面的支持。"]},"en":{"title":"Google proposes ToolGrad: Efficiently generating tool invocation datasets using text 'gradients'","summary":"Google proposes ToolGrad, a data generation framework where an LLM first generates a tool invocation chain and then labels user queries in reverse, published at ACL 2026.","category":"Industry","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google proposes ToolGrad: Efficiently generating tool invocation datasets using text 'gradients' - Aioga AI News","description":"Google proposes ToolGrad, a data generation framework where an LLM first generates a tool invocation chain and then labels user queries in reverse, published at ACL 2026.","url":"https://www.aioga.com/en/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:18.755Z"},"ja":{"title":"GoogleはToolGradを発表：テキストの「勾配」を使って効率的にツール呼び出しデータセットを生成","summary":"GoogleはToolGradを発表しました。これはまずLLMによってツール呼び出しチェーンを生成し、その後ユーザーのクエリを逆方向にラベル付けするデータ生成フレームワークで、ACL 2026で発表されました。","category":"業界動向","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"GoogleはToolGradを発表：テキストの「勾配」を使って効率的にツール呼び出しデータセットを生成 - Aioga AIニュース","description":"GoogleはToolGradを発表しました。これはまずLLMによってツール呼び出しチェーンを生成し、その後ユーザーのクエリを逆方向にラベル付けするデータ生成フレームワークで、ACL 2026で発表されました。","url":"https://www.aioga.com/ja/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:18.785Z"},"ko":{"title":"Google이 ToolGrad 발표: 텍스트 '그라디언트'로 효율적으로 도구 호출 데이터셋 생성","summary":"Google이 ToolGrad를 발표했다. LLM이 먼저 도구 호출 체인을 생성하고, 이후 사용자 쿼리를 역방향으로 라벨링하는 데이터 생성 프레임워크로, ACL 2026에서 발표됨.","category":"업계 동향","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google이 ToolGrad 발표: 텍스트 '그라디언트'로 효율적으로 도구 호출 데이터셋 생성 - Aioga AI 뉴스","description":"Google이 ToolGrad를 발표했다. LLM이 먼저 도구 호출 체인을 생성하고, 이후 사용자 쿼리를 역방향으로 라벨링하는 데이터 생성 프레임워크로, ACL 2026에서 발표됨.","url":"https://www.aioga.com/ko/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:21.158Z"},"es":{"title":"Google presenta ToolGrad: Generación eficiente de conjuntos de datos de llamadas a herramientas usando 'gradientes' de texto","summary":"Google presenta ToolGrad, un marco de generación de datos en el que primero un LLM genera cadenas de llamadas a herramientas y luego se anotan de manera inversa las consultas de los usuarios, publicado en ACL 2026.","category":"Industria","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google presenta ToolGrad: Generación eficiente de conjuntos de datos de llamadas a herramientas usando 'gradientes' de texto - Aioga Noticias de IA","description":"Google presenta ToolGrad, un marco de generación de datos en el que primero un LLM genera cadenas de llamadas a herramientas y luego se anotan de manera inversa las consultas de lo...","url":"https://www.aioga.com/es/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:21.656Z"},"fr":{"title":"Google propose ToolGrad : utiliser le \"gradient\" de texte pour générer efficacement des jeux de données d'appels d'outils","summary":"Google propose ToolGrad, un cadre de génération de données dans lequel une LLM génère d'abord des chaînes d'appels d'outils, puis marque en retour les requêtes des utilisateurs, publié dans ACL 2026.","category":"Industrie","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google propose ToolGrad : utiliser le \"gradient\" de texte pour générer efficacement des jeux de données d'appels d'outils - Aioga Actualités IA","description":"Google propose ToolGrad, un cadre de génération de données dans lequel une LLM génère d'abord des chaînes d'appels d'outils, puis marque en retour les requêtes des utilisateurs, pu...","url":"https://www.aioga.com/fr/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:24.604Z"},"de":{"title":"Google stellt ToolGrad vor: Effiziente Generierung von Tool-Aufruf-Datensätzen mit Text-\"Gradienten\"","summary":"Google stellt ToolGrad vor, einen Rahmen zur Datengenerierung, bei dem zunächst von LLM die Tool-Aufrufketten erzeugt und dann Benutzeranfragen rückwärts gekennzeichnet werden, veröffentlicht auf der ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google stellt ToolGrad vor: Effiziente Generierung von Tool-Aufruf-Datensätzen mit Text-\"Gradienten\" - Aioga KI-News","description":"Google stellt ToolGrad vor, einen Rahmen zur Datengenerierung, bei dem zunächst von LLM die Tool-Aufrufketten erzeugt und dann Benutzeranfragen rückwärts gekennzeichnet werden, ver...","url":"https://www.aioga.com/de/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:23.648Z"},"pt-BR":{"title":"Google propõe ToolGrad: gerar conjuntos de dados de chamada de ferramentas de forma eficiente usando \"gradiente\" de texto","summary":"Google propõe ToolGrad, uma estrutura de geração de dados onde primeiro o LLM gera a cadeia de chamadas de ferramentas e depois rotula inversamente as consultas dos usuários, publicada no ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google propõe ToolGrad: gerar conjuntos de dados de chamada de ferramentas de forma eficiente usando \"gradiente\" de texto - Aioga Notícias de IA","description":"Google propõe ToolGrad, uma estrutura de geração de dados onde primeiro o LLM gera a cadeia de chamadas de ferramentas e depois rotula inversamente as consultas dos usuários, publi...","url":"https://www.aioga.com/pt-BR/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:27.115Z"},"ru":{"title":"Google предложила ToolGrad: эффективно создавать наборы данных для вызова инструментов с помощью текстовых \"градиентов\"","summary":"Google предложила ToolGrad — фреймворк для генерации данных, который сначала создает цепочки вызовов инструментов с помощью LLM, а затем обратно аннотирует пользовательские запросы. Опубликовано на ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google предложила ToolGrad: эффективно создавать наборы данных для вызова инструментов с помощью текстовых \"градиентов\" - Aioga Новости ИИ","description":"Google предложила ToolGrad — фреймворк для генерации данных, который сначала создает цепочки вызовов инструментов с помощью LLM, а затем обратно аннотирует пользовательские запросы...","url":"https://www.aioga.com/ru/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:27.366Z"},"ar":{"title":"جوجل تقدم ToolGrad: استخدام \"التدرج\" النصي لتوليد مجموعات بيانات استدعاء الأدوات بكفاءة","summary":"جوجل تقدم ToolGrad، وهو إطار لتوليد البيانات يتم فيه أولاً إنشاء سلاسل استدعاء الأدوات بواسطة LLM، ثم يتم وسم استفسارات المستخدمين بشكل عكسي، نُشر في ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"جوجل تقدم ToolGrad: استخدام \"التدرج\" النصي لتوليد مجموعات بيانات استدعاء الأدوات بكفاءة - Aioga أخبار الذكاء الاصطناعي","description":"جوجل تقدم ToolGrad، وهو إطار لتوليد البيانات يتم فيه أولاً إنشاء سلاسل استدعاء الأدوات بواسطة LLM، ثم يتم وسم استفسارات المستخدمين بشكل عكسي، نُشر في ACL 2026.","url":"https://www.aioga.com/ar/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:30.707Z"},"hi":{"title":"Google ने प्रस्तुत किया ToolGrad: टेक्स्ट \"ग्रेडिएंट\" का उपयोग करके टूल कॉलिंग डेटा सेट को कुशलतापूर्वक जनरेट करना","summary":"Google ने प्रस्तुत किया ToolGrad, एक डेटा जनरेशन फ्रेमवर्क जिसमें पहले LLM द्वारा टूल कॉलिंग चेन जेनरेट किया जाता है और फिर यूज़र क्वेरी को रिवर्स लेबल किया जाता है, ACL 2026 में प्रकाशित।","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google ने प्रस्तुत किया ToolGrad: टेक्स्ट \"ग्रेडिएंट\" का उपयोग करके टूल कॉलिंग डेटा सेट को कुशलतापूर्वक जनरेट करना - Aioga AI समाचार","description":"Google ने प्रस्तुत किया ToolGrad, एक डेटा जनरेशन फ्रेमवर्क जिसमें पहले LLM द्वारा टूल कॉलिंग चेन जेनरेट किया जाता है और फिर यूज़र क्वेरी को रिवर्स लेबल किया जाता है, ACL 2026 में प...","url":"https://www.aioga.com/hi/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:29.564Z"},"it":{"title":"Google propone ToolGrad: generare efficientemente dataset di chiamata agli strumenti tramite 'gradiente' testuale","summary":"Google propone ToolGrad, un framework di generazione di dati in cui prima un LLM crea catene di chiamata agli strumenti e poi si annotano retroattivamente le query degli utenti, pubblicato in ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google propone ToolGrad: generare efficientemente dataset di chiamata agli strumenti tramite 'gradiente' testuale - Aioga Notizie IA","description":"Google propone ToolGrad, un framework di generazione di dati in cui prima un LLM crea catene di chiamata agli strumenti e poi si annotano retroattivamente le query degli utenti, pu...","url":"https://www.aioga.com/it/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:33.289Z"},"nl":{"title":"Google introduceert ToolGrad: efficiënt genereren van tool-aanroep datasets met tekst 'gradiënt'","summary":"Google introduceert ToolGrad, een datageneratiekader waarbij eerst LLM's tool-aanroepketens genereren en vervolgens gebruikersvragen achterwaarts worden geannoteerd, gepresenteerd op ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google introduceert ToolGrad: efficiënt genereren van tool-aanroep datasets met tekst 'gradiënt' - Aioga AI-nieuws","description":"Google introduceert ToolGrad, een datageneratiekader waarbij eerst LLM's tool-aanroepketens genereren en vervolgens gebruikersvragen achterwaarts worden geannoteerd, gepresenteerd...","url":"https://www.aioga.com/nl/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:33.534Z"},"tr":{"title":"Google, ToolGrad'ı Önerdi: Metin \"Gradyanı\" ile Araç Çağrı Veri Setlerini Verimli Şekilde Üretme","summary":"Google, ToolGrad'ı önerdi; bu, önce LLM tarafından araç çağrı zincirleri üretilen ve ardından kullanıcı sorgularını tersine etiketleyen bir veri üretim çerçevesidir ve ACL 2026'da yayımlandı.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google, ToolGrad'ı Önerdi: Metin \"Gradyanı\" ile Araç Çağrı Veri Setlerini Verimli Şekilde Üretme - Aioga AI Haberleri","description":"Google, ToolGrad'ı önerdi; bu, önce LLM tarafından araç çağrı zincirleri üretilen ve ardından kullanıcı sorgularını tersine etiketleyen bir veri üretim çerçevesidir ve ACL 2026'da...","url":"https://www.aioga.com/tr/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:36.339Z"},"vi":{"title":"Google đưa ra ToolGrad: Sử dụng 'gradient' văn bản để tạo tập dữ liệu gọi công cụ hiệu quả","summary":"Google đưa ra ToolGrad, một khung tạo dữ liệu trước tiên tạo chuỗi gọi công cụ bởi LLM, sau đó đánh dấu ngược các truy vấn của người dùng, được công bố tại ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google đưa ra ToolGrad: Sử dụng 'gradient' văn bản để tạo tập dữ liệu gọi công cụ hiệu quả - Tin tức AI Aioga","description":"Google đưa ra ToolGrad, một khung tạo dữ liệu trước tiên tạo chuỗi gọi công cụ bởi LLM, sau đó đánh dấu ngược các truy vấn của người dùng, được công bố tại ACL 2026.","url":"https://www.aioga.com/vi/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:35.813Z"},"id":{"title":"Google mengajukan ToolGrad: Menggunakan \"gradien\" teks untuk secara efisien menghasilkan dataset pemanggilan alat","summary":"Google mengajukan ToolGrad, sebuah kerangka kerja generasi data di mana LLM pertama-tama menghasilkan rantai pemanggilan alat, lalu menandai balik kueri pengguna, dipublikasikan di ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google mengajukan ToolGrad: Menggunakan \"gradien\" teks untuk secara efisien menghasilkan dataset pemanggilan alat - Berita AI Aioga","description":"Google mengajukan ToolGrad, sebuah kerangka kerja generasi data di mana LLM pertama-tama menghasilkan rantai pemanggilan alat, lalu menandai balik kueri pengguna, dipublikasikan di...","url":"https://www.aioga.com/id/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:38.790Z"},"th":{"title":"Google เสนอ ToolGrad: ใช้ \"กราดิเอนต์\" ของข้อความเพื่อสร้างชุดข้อมูลเรียกใช้งานเครื่องมืออย่างมีประสิทธิภาพ","summary":"Google เสนอ ToolGrad กรอบการสร้างข้อมูลที่เริ่มจาก LLM สร้างโซ่การเรียกใช้งานเครื่องมือ จากนั้นย้อนกลับมาทำการติดป้ายคำค้นหาของผู้ใช้ เผยแพร่ใน ACL 2026","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google เสนอ ToolGrad: ใช้ \"กราดิเอนต์\" ของข้อความเพื่อสร้างชุดข้อมูลเรียกใช้งานเครื่องมืออย่างมีประสิทธิภาพ - ข่าว AI Aioga","description":"Google เสนอ ToolGrad กรอบการสร้างข้อมูลที่เริ่มจาก LLM สร้างโซ่การเรียกใช้งานเครื่องมือ จากนั้นย้อนกลับมาทำการติดป้ายคำค้นหาของผู้ใช้ เผยแพร่ใน ACL 2026","url":"https://www.aioga.com/th/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:39.130Z"},"pl":{"title":"Google przedstawia ToolGrad: efektywne generowanie zestawów danych wywołań narzędzi za pomocą „gradientu” tekstu","summary":"Google przedstawia ToolGrad, ramy generowania danych, w których najpierw LLM generuje łańcuch wywołań narzędzi, a następnie odwraca etykietowanie zapytań użytkowników, opublikowane na ACL 2026.","category":"行业动态","source":"Google Research：Blog（网页）","aggregationSource":"Google Research：Blog（网页）","pageTitle":"Google przedstawia ToolGrad: efektywne generowanie zestawów danych wywołań narzędzi za pomocą „gradientu” tekstu - Aioga Wiadomości AI","description":"Google przedstawia ToolGrad, ramy generowania danych, w których najpierw LLM generuje łańcuch wywołań narzędzi, a następnie odwraca etykietowanie zapytań użytkowników, opublikowane...","url":"https://www.aioga.com/pl/news/cmtw5rpj204syrolksxq1yofe/","contentTranslated":true,"sourceHash":"ba1ca9f6053d90f8","translatedAt":"2026-09-11T00:01:41.908Z"}},"evidenceTier":"verified-news","reviewStatus":"automated-ingest","indexable":true,"editorialCover":"/page-visuals/topic-timeline.png"}}