Read on Twitter
Funny fact: when I was a software developer, I thought that #DevOps is a challenging practice; now I develop and advocate #MLOps and sometimes feel that it would be so much easier to be a #DevOps 
Here comes a thread about differences between #MLOps and #DevOps.
Here comes a thread about differences between #MLOps and #DevOps.
In #DevOps, the state of your application is defined by a commit in your source code (or a bunch of commits if you have microservices). In #MLOps, you also have a version of your dataset (or datasets) to manage, which is not that simple because you can't put those 74GB in Git.
In #DevOps, building your project takes up to several minutes. In #MLOps, you train instead of build, which may take up to several days. Because of that, the classic CI/CD approach simply doesn't work.
In #DevOps, building a project is very cheap when it comes to compute. In #MLOps, running one training cycle can cost you up to several thousand dollars due to the expensive GPU compute many modern models need.
In #DevOps, the resulting software is usually deterministic: you know the exact output for given inputs, which makes your software (relatively) easy to test. In #MLOps, models are non-deterministic, so that you can't check exact outputs for given inputs, only some metrics.
In 
