I have always been a staunch practitioner and "advocate" of locally deploying large models. Over the past year, my articles have frequently featured practices in local deployment: from single-machine to cluster deployment, from one-click setups to modifying source code and recompiling, and from Quickstart guides to network optimization.
However, I have always known that this path is only suitable for a few. For many "knowledge workers," the entry barrier for locally deploying a large model is actually lower than installing a AAA game on a computer.
Yet, for the vast majority of people, the barrier for daily use is higher than any task they have faced before.
Thanks to DS (DeepSeek), there has been a nationwide surge in learning. However, the journey from starting to giving up might take less than a week. I have attempted to establish some simple criteria; roughly speaking, failing to meet any of these is a reason to cut your losses:
- Do you possess a local database or knowledge base?
- Is a significant portion (over 5%?) of that database or knowledge base self-generated?
- Do you have truly sensitive data that you do not want model providers to access?
- Do you need to process more than a hundred local data points or pieces of knowledge every day?
- Do you clearly know what form these data points should be processed into?
- Have you found a fixed UI (application) where you can complete over 90% of your work?
- Do you possess basic knowledge of operating systems?
- Are you willing to spend two, three, or even more days solving a system or program issue?
- Have you already decided what your final product will look like?
- Can you endure the deep sense of frustration when the content you produced with great effort is far inferior in quality to the "one-click generation" of commercial models? And then, can you pull yourself together and keep going?
If you match any of these (meaning you fail to meet a requirement), please give up promptly and continue to try the best commercial models. Do not let an "invisible wall" that doesn't suit you block your view of the sky.