llm_trainer
收藏资源简介：
优点： （1）国内可用，容易访问，社区活跃度强，数据集丰富； （2）60小时的免费GPU使用，CPU不限量,但是单次使用GPU最多8小时； （3）安装TF和Pytorch比较方便和容易，不像某度只能用PP 缺点： （1）60小时的时间有点短。。。 （2）新环境的notebook上传数据后，下次会丢失，老版本不会，就是这个： 新老版本的对比 总结：这是目前用过的很好用的能够方便访问的免费GPU资源，谢谢阿里 2、移动九天平台（推荐指数：⭐️⭐️⭐️）地址：https://jiutian.10086.cn/edu/#/home 移动推出的集比赛、数据、训练于一体的平台，貌似不是很活跃，数据集都没多少，但是模型训练的羊毛还是要薅的！ 移动九天 （别问为什么是灰色的界面，但凡看点新闻都知道，在这里缅怀一下伟人！） 优点： （1）免翻访问，TF和Pytorch都有； （2）可以免费上传数据集，容量大200G； （3）签到、邀请可以增加训练时长，按照算例豆来计算，具体可看官网； 缺点： （1）安装环境出错，我安装许久的Pytorch环境，失败了。。。。。可能方法没对 （2）算例豆消耗特别快，要主要计算自己的模型训练时间 总结：可以一试，毕竟也是羊毛 最后： 以上4个是我使用过的免费GPU资源，总体来说：我的体验是阿里云>Colab>SageMaker>移动九天，希望今天的文章能够对你有所帮助，喜欢的话，三连支持一下，老铁们！ 作者：马克莱报到 https://www.bilibili.com/read/cv20275242/ 出处：bilibili
First Platform (presumably Alibaba Cloud's machine learning platform): ### Advantages: 1. Accessible within mainland China, easy to obtain, with an active community and abundant dataset resources. 2. Offers 60 hours of free GPU usage quota, with unlimited CPU resources, while each GPU usage session is limited to a maximum of 8 hours. 3. Easy to install TensorFlow and PyTorch, unlike Baidu which only supports PaddlePaddle. ### Disadvantages: 1. The 60-hour quota is relatively limited. 2. Data uploaded to notebooks in the new environment will be lost upon next access, while this issue does not affect the old version. #### Comparison between New and Old Versions: Summary: This is one of the most conveniently accessible free GPU resources I have ever used. Kudos to Alibaba Cloud. --- 2. Mobile Jiutian Platform (Recommendation Rating: ⭐️⭐️⭐️), URL: https://jiutian.10086.cn/edu/#/home This is a platform launched by China Mobile integrating competition hosting, dataset provision and model training. It appears not very active, with a relatively small number of available datasets, but it is still worth utilizing the free training resources! > Mobile Jiutian > (Don't ask why the interface is gray. Anyone who follows current affairs will understand the reason; let's pay tribute to the great man here!) ### Advantages: 1. No VPN required for access, with support for both TensorFlow and PyTorch. 2. Allows free dataset uploads with a maximum capacity of 200 GB. 3. Training duration can be extended through sign-in and invitation activities, calculated using "Calculation Beans"; please refer to the official website for specific rules. ### Disadvantages: 1. Environment installation often encounters errors. I spent a long time trying to set up a PyTorch environment but failed... likely due to incorrect configuration methods. 2. The consumption of "Calculation Beans" is quite rapid, so please carefully estimate your model training time. Summary: Worth a try, as it also provides free resources. --- Final Note: The above are 4 free GPU resources I have personally used. Overall, my experience-based ranking is: Alibaba Cloud > Colab > SageMaker > Mobile Jiutian. I hope this article is helpful to you. If you find it useful, please support with likes, favorites and reposts, everyone! Author: Mark Laibao https://www.bilibili.com/read/cv20275242/ Source: bilibili
