
LlamaCon: New API, Tools, and Protections Announced
LlamaCon unveiled the Llama API with fine-tuning, faster inference via Cerebras/Groq, Llama Stack integrations, enhanced security tools, and Impact Grants, reaching one billion downloads.
LlamaCon Highlights: Exciting New Tools and Initiatives
Here's a rundown of the key announcements from LlamaCon:
-
Llama API (Limited Preview): Combines the best of closed models with open-source flexibility. Offers easy API key creation and interactive playgrounds to explore Llama models (including Llama 4 Scout and Llama 4 Maverick). Includes lightweight SDKs in Python and Typescript and is compatible with the OpenAI SDK.
-
Fine-tuning and Evaluation Tools: Tune custom versions of the Llama 3.3 8B model within the API. Generate data, train, and evaluate your model's quality. Your data isn't used to train Meta's AI models, and you retain control of the models you build.
-
Faster Inference: Collaboration with Cerebras and Groq provides faster inference speeds using the Llama API. Early experimental access to Llama 4 models powered by Cerebras and Groq is available by request.
-
Llama Stack Integrations: Expanding collaborations, including an integration with NVIDIA NeMo microservices. Working with partners like IBM, Red Hat, and Dell Technologies on new integrations.
-
New Llama Protections: Releasing new tools including Llama Guard 4, LlamaFirewall, and Llama Prompt Guard 2. Updates to CyberSecEval 4 for evaluating AI system security.
-
Llama Defenders Program: Select trusted partners can access AI-enabled tools to evaluate system security against potential threats.
-
Llama Impact Grants: Announcing 10 international recipients of the second Llama Impact Grants, totaling over $1.5 million USD. Supports companies, startups, and universities using Llama for transformative change. Examples include E.E.R.S. (US), Doses AI (UK), Solo Tech (US), and FoondaMate (Africa).
-
One Billion Downloads: Llama has surpassed one billion downloads, establishing itself as a leader in the open-source AI ecosystem