In the tests carried out by the Spanish – the free android to test the AI DEPEEK R1 in a laptop appeared a program that is becoming the key to install different major language models and thus make consultations to its chatbots with the safety of that all Messages will be maintained locallyensuring your privacy.
Ollana is a free tool or program That can be installed both in Windows and a Mac as well as GNU / Linux, so there are almost no limits to download any language model greater than the PC and thus do the relevant tests.
You have to take a very point into account and that is that the abilities of an AI are always teams subject to the team’s hardwarewith the exception of Deepseek R1 that, being so efficient, can even be installed on an Android mobile (although always with its limitations in the lentidud of the answers, but it can be said that it is a first and big step).
The variety of ‘distilled’ models of Deepseek R1 That they have been launched these days allow more flexibility when installing this major language model, although the base model, due to its magnitude and demand for greater resources, would need a system that, even with a gaming of very high specifications, It would be more complex to install.
But he does not mean that some of his submodels, such as versions 1.5 B, 8 B, 14b And others, allow a user with a good Mac or a PC with a powerful GPU (if one of the last 5090 or the previous RTX 4070, 4080 or 4090, better than better), you can have a chatbot on a local chatbot which shows his “deep thought” in the answers and has been able to overcome GPT-4O in some Benchmarks.
Consult Deepseek AI
The free android
The efficiency of these Deepseek models is such that even Microsoft has advanced to the rest of the technological giants by announcing yesterday that will deploy 1.5b distilled R1 in their co -pilot+ PCs and soon the submodels 7b and 14b distille to take advantage of their capabilities.
Apart from being able to install the major language model, there is another great virtue and it is that an Internet connection is not needed To consult the chatbot. As well as the censorship that could be attributed completely disappears.
The free android
Another important point is that all communication with the chatbot will be done through the system symbol in Windows or the terminal on the PC. There are alternatives to use the major language model as it would be Chatbox AIa desktop interface Open Source which also allows you to download the data locally in Mac and Windows.
How to download and use ollama
The first thing is to go to the Ollama website for download the application On the computer, either a Mac or a PC with Windows or Linux. It starts and will be installed in a simple way as can be done with another program in these systems.
Executing the system symbol command
The free android
Now it’s time to open system symbol on the PC To be able to operate with Ollama from the search engine in Windows to type its name (it is recommended to use administrator rights). CMD opens and this series of commands are directly typed according to the model you want (and having in mind to try one and others according to the hardware):
- 1.5B version (the smallest): Ollama Run Deepseek-R1: 1.5b
- Version 8b: Ollama Run Deepseek-R1: 8b
- Version 14b: Ollama Run Deepseek-R1: 14b
- 32b version: Ollama Run Deepseek-R1: 32b
- 70b version: Ollama Run Deepseek-R1: 70b
Once the chosen model is loadedthe following will be to communicate directly as it would be done with its web or chatgpt version (it is identified with >>>). And it is mentioned again that there is no interface that reminds what is used in its apps, it is all as ‘without filters’, but does not mean that it is equally useful.
The ‘deep thought’ will appear Without any type of label in each response and so you see how AI thinks. When you begin to answer in Spanish it will be known that the final answer will be giving the consultation. Another important point: When the consultations are over, it is recommended to close ollama so as not to have it loaded in the system.
From Ollama you can look for more greater language models As call3.3, Phi4 and many others and thus test, but the one that has demonstrated great efficacy and less exisgence in resources is the aforementioned, just 8b. And what has been said, if there is a larger hardware, you can try another more effective model in your answers and in the speed that makes them, this being a very important point.
