Deepseek can be installed on a PC or Mac locally so that all requests made remain in the machine. This is another of the experiences that has left the speechless world by not asking as many resources and providing the ability of GPT-4O in a “distilled” model which offers great flexibility in its use. You can also mobile Like Samsung Galaxy S24 Ultra itself.
As is the case when Deepseek is installed on a PC or Mac, depending on the main linguistic model used, its effectiveness will be more or less, not only in the answers you can give to consultations, but in the speed of speed to develop them.
There is another important point in Depseek models which can be loaded on a high mobile, and is that Optimized versions of Deepseek R1 Their purpose is to be installed in machines with fewer resources.
And better the PC or the mobile, it will offer a better experience, so you must try to choose the model that will be well installed, because if this is not done properly, the speed of your responses can be considerably reduced as well as your precision.
The tests carried out by the Spanish – the free Android goes almost more through an experience, since the model which has been able to offer more precise answers to Need more time to treat them (As with Depseek-R1-Distill-Qwen-14b-iq2_m) and being Deepseek R1, he will show in the text all the logic of each of his “deep thoughts” to give the final answer.
Ask for the Torrijas recipe in Depseek in a Samsung Galaxy S24 Ultra
The free android
It is really a Sample of the effectiveness of this major language model This competes with GPT-4O and can be installed on an Android mobile. Something practically unthinkable there are days and which is already available in our hands, although in a limited way.
The Deepseek code repository of its different models is available from FaceA platform that becomes one of the nodes to share tools, models and sets of data related to generative artificial intelligence.
Necessary requirements
The compilations we will use were published just a few days ago As happened with Deepseek R1, which is also available from this repository, although it needs a large number of resources that can offer a PC, even if it is the best game.
El Subodelo Deepseek-R1-Distill-Qwen-14b-iq2_M He needed 724 ms per token with 1.38 tokens per second to give all the details on how to make a toast recipe and with certain incorrect words. It's slow, but it is also a question of trying other submodels to speed up the answer and check that their answers are more or less precise.
Samsung mobile is one of the best current mobiles for its specifications and certain requirements are recommended to execute some of the Depseek R1 models which can be downloaded from the Hugging face:
- Chip: Snapdragon 8 Elite, Snapdragon 8 Gen 4 / Gen 3 and some Snapdragon 7+ Gen 3 as well as those of Mediatek as Dimensity 8400 or similar.
- RAM memory: At least 8 GB for 8B / 14B models of Deepseek.
- Storage: At least 12 GB for free.
How to install Depseek on a mobile
To install Depseek on the mobile Installation of PocketPal AI, an application is necessary open source This facilitates all steps to download a major linguistic model, then load it to work with the chatbot.
He is downloaded From the Google Play Store on Android and is also available for iOS on the iPhone from the App Store. Once the application is open, just click on the “Go to the models” button, then touch “+” and choose “Add from the grip face”.
Deepseek R1 submodels
The free android
Here, the full repository of this platform will appearSo the search engine must be used to type “Depseek”. Among the whole list of results, it is recommended to use a wider model such as 8b, because 1.5b or 7b are faster, but more imprecise in their responses.
Each model then has its submodels which vary in file capacity and size to download. The larger the resources, the larger the resources, so depending on the mobile, we will have to prove one or the other, we will use Deepseek-R1-Distill-Qwen-7b-Q3_K_L with 4.09 GB and 7.62B parameters.
The model to install and the adjustment to be made
The free android
We download the submodèle and at the end, we go to the “Models” section In the main menu. It is pressed on “Settings” in the downloaded model and on the next screen, you must configure “N predict” with 4096 tokens to properly generate text.
He returns to the previous screen and is played in “LOAD” To start the selected model and so that you can go to a conversation window in which you can consult, see the answers and even start another conversation with the AI.
Deepseek will return the answers by “deep thought” marking it as “thinking …”. When you finish, he will give the answer to continue with the conversation. We remember again that once the session is over, it is recommended to download with “unloading” on the model of the previous screen.
