包阅导读总结
1. 关键词：GitHub Models、AI 模型、Azure AI、开发人员、集成
2. 总结：GitHub 推出了免费的 GitHub Models，让开发者能在其工具生态系统内探索多种 AI 模型，提供测试提示和调参的 playground，与 GitHub Codespaces 和 CLI 集成，目前处于封闭预览阶段，支持多种模型但不支持用户上传自定义模型。
3. 主要内容：
– GitHub 推出 GitHub Models
– 目标是让开发者在 GitHub 工具生态中探索 AI 模型，便于使用 Azure AI 部署服务
– 包括公私模型，处于封闭预览
– 功能特点
– 提供 playground 测试提示和调参
– 可在 Marketplace 选择模型，查看详情和限制
– 方便切换模型对比行为
– 结果满意可部署到 Azure AI
– 集成情况
– 与 GitHub Codespaces 集成，可从 playground 启动预配置环境
– 与 GitHub CLI 集成，在终端运行提示和查看结果
– 优势与不足
– 优势：无 Azure AI Studio 企业级安全约束
– 不足：付费仅能通过升级到 Azure AI
– 支持与不支持的模型
– 支持 Llama 3.1 等多种模型
– 不支持用户上传自定义模型
思维导图：
文章来源：infoq.com
作者：Sergio De Simone
发布时间：2024/8/9 0:00
语言：英文
总字数：485字
预计阅读时间：2分钟
评分：87分
标签：AI 模型,GitHub,Azure AI,开发者工具,AI 探索平台
以下为原文内容
本内容来源于用户推荐转载，旨在分享知识与观点，如有侵权请联系删除 联系邮箱 media@ilingban.com
GitHub has launched GitHub Models, a free capability aimed at letting developers explore various AI models from within the GitHub tool ecosystem and make it easier to deploy AI-based services using Azure AI. GitHub Models includes both private and public models and is currently in closed preview.
GitHub Models provides a playground that lets developers test prompts and fine-tune model parameters. In order to use a model, you go to the GitHub Marketplace and select the one you want to work with, for example Meta-Llama-3-70B-Instruct. From the model page, you can get further details about how toget started, as well as inspect rate limits associated to that model. The playground also makes it easy to switch from one model to another and compare their behavior. For example, GitHub describes a scenario where a developer defines their prompts using GPT4o, then switches to a lighter-weight model like Phi-3 mini to see how it performs with those same prompts. Finally, when they are satisfiedwith the results, the model can be deployed directly to AzureAIby replacing GitHub tokens withAzure credentials.
The GitHub Models playground is also integrated with GitHub Codespaces, allowing developers to spin a new pre-configured Codespace environment directly from the playground once they are satisfied with the results. The integration with Codespaces appears to rely on existing SDKs that are already available for Azure AI with support for several languages. The generated codespace uses the entitlements that come with a GitHub account, which means you do not need to add an API key or sign up for any other services, making it straightforward to start coding with any of the supported AI models.
The new capability is also integrated with the GitHub CLI, allowing users to run a given prompt against one of the supported models and see the results in the terminal. The run command can be combined with other git commands, for example, to summarize the last commits from a repo, you could use:
git log -n 10 | gh models run phi-3-mini-4k-instruct "Summarize this commit history. Use max 1 line per commit"A few commenters on Hacker News pointed out that while closely related to Azure AI Studio, GitHub Models has the advantage of being free from the constraints deriving from Azure AI Studio enterprise-level security features, which makes the latter cumbersome to use when it comes to authentication and lagging behind competitor playgrounds in terms of supported models versions. On the other hand, somecriticize that the only paid tier available is through upgrading to Azure AI.
At the time of the announcements, GitHub Models supports Llama 3.1, GPT-4o and GPT-4o mini, Phi 3, Mistral Large 2, and others. Not supported is the possibility for users to upload and run their custom models. As a final note, according to GitHub, no prompts or generated outputs are shared with model providers, nor used to train or improve the models.
