Google has announced its new AI model Gemma 2B, a new open-source AI model under the Gemma 2 banner. Gemma 2B is a compact but more efficient language model designed to perform tasks such as text generation, translation and answering questions. Google claims that despite its small size, Gemma 2B outperforms its larger competitors including ChatGPT 3.5 in terms of speed and conversational capabilities. Along with Gemma 2B Google has also introduced new AI tools, ShieldGemma and Gemma Scope which are designed to enhance AI capabilities while prioritising security and transparency.

The new AI model Gemma 2B is the highlight of Google’s recent announcement as Google says that despite having only 2 billion parameters, it outperforms many larger models in the GPT-3.5 category. Apart from speed, Gemma 2B is also said to offer impressive flexibility. According to Google the model can run efficiently on a wide range of hardware, from edge devices to powerful cloud platforms. And this versatility will allow the model to be accessible to a wide range of applications, including educational tools, personal assistants, and more.

Google has optimized the model for NVIDIA GPUs, making it accessible to the wider developer community. Additionally, it is compatible with popular AI frameworks such as Keras, JAX, Hugging Face, and others, simplifying integration into existing projects.

Meanwhile, Google has also introduced ShieldGemma which is a collection of security classifiers designed to help developers identify and filter harmful content within AI systems. The new tools will help users detect and prevent the spread of hate speech, harassment, sexually explicit content, and other harmful materials. Google notes that the open nature of these classifiers promotes transparency and encourages collaboration within the AI ​​community, which aligns with Google’s broader efforts to set industry standards for safe AI deployment.

In addition, to help researchers understand how AI models like Gemma 2 work, Google has also launched a new tool called Gemma Scope. According to Google, this new tool uses sparse autoencoders to provide users with insights into the inner workings of the Gemma 2 model. By breaking down complex information into more simple understandable formats, Gemma Scope will allow researchers to study how AI models make decisions, helping them develop reliable and accountable systems.

For the availability of new models and tools, Google has decided to provide users with free access to these Gemma 2 models, ShieldGemma, and Gemma Scope. The Gemma 2 2B model weights can be downloaded from platforms such as Kaggle, Hugging Face, and Vertex AI Model Garden. Additionally, the models can be tested on Google AI Studio and Google Colab, making it accessible even to those with limited resources.