All About Google Gemma

On Feb 21, 2024, Google launched Gemma which is a pre-trained, fast, and lightweight model that can easily run on laptops. Now, ChatGPT, Gemini, and many other LLM models are also present. But the problem is they are a product of some company. Individual students or developers can’t use them. So, Google launched a series of open models that developers and researchers can use.

As per Google, Gemma is a lightweight and open model that anyone can use. This model was built from the same research and technology that was conducted and used to create Gemini. Google mentioned in their blog that:

1. This model is released in two different sizes. Every size has pre-trained and instruction-tuned variants.
2. They have also created a toolkit that will help in creating safer AI apps using GEMMA.
3. There is a toolchain for inference and supervised fine-tuning (SFT) in all major frameworks.
4. Gemma will be available to use via Kaggle, Colab, Hugging Face, and Google’s Vertex AI.

To be honest, I think Google did a pretty amazing job by releasing Google Gemma. This is because Google Gemini is a closed AI model. But, Gemma is an open-source model.

Moreover, you can also fine-tune the Gemma model. There are various tools and systems that Gemma supports. You can make multi-framework tools. You can also optimize for Google Cloud. What’s even better? There is a cross-device compatibility. There are free credits that you can use for research and development. You can also apply for Google Cloud credit — if you have one.

Now, people are concerned about misuse of personal information — well, that’s a genuine issue. Fortunately, Google has clarified that personal information was filtered during the pre-training of Gemma. Moreover, extensive fine-tuning and reinforcement learning from human feedback (RLHF) was also done to align models with responsible behaviors. To make sure that developers and researchers use GEMMA to build only safe and responsible AI apps, the toolkit includes a method that can be used to build safety classifiers. There is a debugging tool as well.

By the way, Gemma means precious stone.