Modele i wagi AI38
Jak wybrać: Modele AI
This page covers the large models themselves and the platforms for getting and running them: open-source models, model communities, local runtimes and hosted model services. It suits people who want to deploy models themselves, compare models, or use AI offline.
What's on this page
- Model communities and downloads: Hugging Face, the largest community for models and datasets, and ModelScope, Alibaba's Chinese model community.
- Open models such as Meta's LLaMA and Google's Gemma, which you can download, run on your own hardware and fine-tune for free.
- Local runtimes: Ollama runs open models on your machine with a single command; Chatbox AI, Cherry Studio and AnythingLLM are desktop clients that connect to local models and provider APIs.
- Hosted inference: Replicate runs open models in the cloud through an API, with no GPU of your own.
Local or cloud
Running locally keeps data on your computer and has no per-call cost, but needs enough GPU memory — large models may not run, or run slowly. Small local models suit everyday use and sensitive material; use cloud APIs when you need the most capable models.
Further reading
This site has a guide to running large models locally and detailed articles on Ollama, Gemma and vLLM.
Odkrywaj dalej możliwości AIPrzeglądaj wszystkie narzędzia