Llama 3.2: On-device 1B/3B and Multimodal 11B/90B Models – Access via API ??

Welcome to the?latest AI in 5?newsletter with Clarifai!

Every week we bring you new models, tools, and tips to build production-ready AI!

Here's a summary of what we will be covering this week: ??

• New models: Llama 3.2 from Meta
• Notebook: Fine-tuning image classification models using Python SDK
• Watch the webinar: AI for marketers - Content generation and personalization with visual search
• Tip of the Week: Inference with Multimodal Inputs on the Llama 3.2 11B Vision Instruct Model

Llama 3.2 from Meta ??

Meta released Llama 3.2 yesterday, featuring small and medium-sized multimodal LLMs (11B and 90B) as well as lightweight text-only models (1B and 3B) designed for mobile and edge devices.

Both the base and instruction-tuned versions have been open-sourced by Meta.

Key Highlights:

1. Vision Use Cases - The multimodal Llama 3.2 models support image reasoning tasks, such as document-level understanding (including charts and graphs), image captioning, and more.
2. Performance - The 3B model outperforms the Gemma 2 (2.6B) and Phi 3.5-mini models in tasks such as instruction-following, summarization, and tool use, while the 1B model is competitive with Gemma.
3. SLMs Capabilities - With a 128k-token context window, the lightweight 1B and 3B models excel at multilingual text generation and tool-calling tasks. They are ideal for building personalized, on-device agentic applications where data remains on the device.

The Llama 3.2 11B Vision Instruct and 3B Instruct models are now available on the Clarifai platform. Try them out and access them via API!

Fine-tuning Image Classification models using Python SDK ??

You can use pre-trained models from the Clarifai community as a starting point. However, in many cases, accuracy and the ability to tailor solutions precisely outweigh speed and ease of use.?

Additionally, you may need a model to learn new features not recognized by existing models. In these cases, you can "deep fine-tune" your custom models and integrate them directly into your workflows.?

You can perform deep training within the platform. Check out the guide here, but the notebook below provides a tutorial on model training for the visual classifier model type using the MMClassification_EfficientNet template with the Python SDK.?

Fine-tuning Image Classification models

Content Generation and Personalization with Visual Search ?

Yesterday, we hosted a webinar: Content Generation and Personalization with Visual Search.?

If you missed it, don’t worry! You can access the recording below to learn more on:

• Why and how to use AI for personalized content and product recommendations
• How to organize content by tagging and enriching data, and leverage AI models to create personalized content strategies
• How to create a common language and framework for AI development to enhance marketing strategies

Find the recording below ??

Content Generation and Personalization with Visual Search

Tip of the Week: Inference with Multi-Modal Inputs ??

Inference with Multi-modal inputs using Clarifai SDK.?

Multi-modal inputs refer to feeding multiple types of data into a single model for processing and analysis.

Below is an example of sending both image and text inputs to the latest Llama 3.2 11B Vision Instruct model, showcasing the versatility of the Predict API in handling diverse data types.?

Check out the complete guide here.

Want to learn more from Clarifai? “Subscribe” to make sure you don’t miss the latest news, tutorials, educational materials, and tips. Thanks for reading!