TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

MLops Startups Profits

1 点作者 spicyramen超过 3 年前
I have seen the rise of MLOps startups such as Cohere.ai, Saturn Cloud, Deepnote, octoML, weights and biases, etc. How really each of them are using VC money and making profits or just burning money? Reasoning: I work in ML and computer resources are extremely expensive hence making a business out of it doesn't make much sense in the long run

1 comment

maininformer超过 3 年前
I am using such a company for my day to day, and in my experience it is less about providing resources and more about managing everything on them.<p>Model versioning, training artifacts versioning, training code versioning, training and test data versioning, providing developer environment on a gpu machine, serving a model and some more tasks are very costly for a company to implement on their own using open source tools.<p>A manager platform that does all that certainly is valuable and I will recommend it to any new ML team.<p>If these use cases don&#x27;t click with you I suggest thinking through working with ML models long term on a team with a big customer base. What if you introduce new data and your model does better but then later does worse than the original? what if you tested your model and it had great performance but after 3 months in production it sucks, now you want to go back to your test data at the time and see why. what if you personal machine does not have gpus? what if you need a custom dev environment? what if different customers need your model at different versions of your training data.<p>I think it is rightfully a nascent niche.
评论 #29221507 未加载