In Short
- Google released an open-source model called Gemma in two sizes. The 2B model weighs just 1.5GB so you can run it on your low-resource computer locally.
- The model consumes close to 1.4GB RAM and generates 10 tokens per second on a 10th-gen Intel i3-powered PC.
- You can use the model for creative tasks in English including text generation, summarization, and basic reasoning.
Just recently, Google introduced its first open-source AI model called Gemma in two sizes, Gemma 2B and 7B. The model is decent for creative tasks in English like text generation, summarization, basic reasoning, etc. Due to its small size, the model can be downloaded and locally used on a low-resource computer without an internet connection. So here is our tutorial to download and run the Google Gemma AI model on Windows, macOS, and Linux.
Table of Contents
Download Google Gemma 2B Model
- Go ahead and download LM Studio (Free) on your computer. Next, launch the program.
- On the homepage, you will find the “Google’s Gemma 2B Instruct” model. You can also manually search for “Gemma”. Now, open it.
- Next, simply click on “Download”. It’s a 1.5GB file as the Gemma 2B model has been 4-bit quantized to compress the model size and reduce memory usage. If you have 8+ GB RAM, you can download the 8-bit quantized model (2.67GB) that will offer better performance.
Recommended Articles
Run Google Gemma Offline Using LM Studio
- In LM Studio, move to the “Chat” window on the left side.
- After that, click on “Select a model to load” and choose “gemma”. It will consume about 1.4GB RAM.
- Now, go ahead and ask your questions. In my experience, the tiny 2B Gemma model is good for creative tasks in the English language. It refuses to answer anything of value where opinion on any subject is sought (e.g. is C better than Rust?).
- You can also set a system prompt on the right to customize the behavior of the AI model.
So this is how you can download and run Google’s Gemma model on your PC, Mac, and Linux computer. While I find Google’s open-source model quite bland, it might be useful for people who are looking for a simple AI model to run locally.
Recommended Articles
Anyway, that is all from us. If you are looking for a true local AI assistant, I would strongly recommend checking out Open Interpreter. It works like ChatGPT’s Code Interpreter and performs actual tasks on your computer. Finally, if you have any questions, let us know in the comment section below.
Sagnik Das Gupta
View quick summary
Palmsy is a new app in the block that aims to be a social media detox app by 'forging' the experience. It does so by faking the likes you get on your posts on the platform. Super strange and unique, right? That's what got me too, and I just had to give it a try. I will say this, the app certainly has its own weird way to getting you to validate yourself.
Anshuman Jain
View quick summary
YouTube is exploring the idea of displaying ads on paused videos, following positive results from initial tests on smart TVs. The ads will pop-up when you pause a video shrinking the video down and you will have to dismiss the ad to continue playing the content you were watching. This could be a move to drive more users to YouTube Premium.
Kanika Gogia
View quick summary
The Photos app on an iPhone automatically scans your photos and recognises the faces of people and pets. To give them a name, go to Photos app -> Albums -> People, Pets & Places, tap on the photo and then tap on Add Name.
You can also tap your pets and people manually. For this, go to the photo, swipe it up, and then tap on the pet or person with a question mark.
Abubakar Mohammed
View quick summary
Samsung's next Galaxy Unpacked event is reportedly set for July 10th in Paris. This aligns with Samsung's usual launch schedule for its second Unpacked event in the mid-year. As per reports, the company is expected to unveil the Galaxy Z Fold 6, Z Flip 6, alongside the Galaxy Watch 7 series with a potential debut for the Galaxy Ring.
Anshuman Jain
Abubakar Mohammed
View quick summary
In this article, we take a look back at the history of Nokia as the different times the brand died, was given a new life and then died again. From refusing to adopt the touch screen to switching to Windows Phone and finally making the jump to Android with HMD. This story of Nokia is a brief lesson on what not to do as a smartphone manufacturer.
Anmol Sachdeva
View quick summary
iOS developer Adam Demasi has been successful in tricking his iPhone into believing that he is in Italy while living in Australia. He did this to get his hands on EU-exclusive iPhone features like third-party app stores, option to choose the default browser and direct app downloads from websites. However, the process is complex and not feasible for everyone to try.
Arjun Sha
View quick summary
We have extensively compared the Snapdragon X Elite and Apple M3 chipsets. In terms of CPU performance, both processors are on par, however, Apple achieves identical performance at a lower frequency. On the GPU front, the Adreno GPU on X Elite can't beat the Apple M3 GPU. But it does rival the older M2 GPU. Finally, on the NPU front, Qualcomm takes the lead with the powerful Hexagon NPU.
Load More
