Get free access to the Brave Search API throughout the hackathon. Brave’s API is powered by our independent search index, and offers:

• Access to an index of 20 billion pages in a single call
• Tens of millions of pages added every day
• Easy, intuitive data structure
• Data populated by real human interactions, not crawlers

Our API is a perfect complement to Mistral models to create AI chatbots, train models, and more. Here are just a few example uses:

• Web search for getting real-time queries on current events
• Extra alternate snippets for rounding out prompt context
• Discussions on forums like Reddit for more natural conversation about topics
• Image search for adding visual responses to your chatbot
• And much more, get creative!

To view API documentation, create a free account at https://api.search.brave.com/register

API Key: TBD

• 10 queries per second limit up to 100,000 queries shared by all participants using the key. Please be mindful of other hackers.

• Phoenix library: https://github.com/Arize-ai/phoenix
• Use case tutorials: https://docs.arize.com/phoenix/notebooks
• LLM tracing demo: https://phoenix-demo.arize.com/projects/UHJvamVjdDow
• Embeddings analyzer demo: https://phoenix-demo.arize.com/model/embeddings/RW1iZWRkaW5nRGltZW5zaW9uOjA=

You will be invited to the https://huggingface.co/CV-Mistral-Hackathon workspace upon entry. If you have not been invited, please email [email protected]

When using hf.co to create Spaces with hardware upgrades, deploy Inference Endpoints, or training using Autotrain-Advanced: Endpoints are billed by uptime; our documentation on endpoints can be found here.

• You can create / manage endpoints at https://ui.endpoints.huggingface.co/endpoints
• In order to stop incurring cost on running endpoints, you’ll need to delete or pause them
• We also have an automatic scaling to 0 feature!

Hardware upgrades for Spaces are billed by uptime, regardless of whether you are using the app or not; our documentation on spaces can be found here. It’s recommended to either:

• Downgrade the Space to free hardware when you are done using it
• Set up a "sleep time" in the repo settings so that the Space automatically gets downgraded if not used
• Or pause a Space, so that the Space is on hold and will not use resources until manually restarted by the owner of the Space

All documentation can be found at https://huggingface.co/docs.

Fireworks.ai provides a fast, reliable platform for developers to run generative artificial intelligence (AI) models at scale. Our mission is to democratize AI for developers and businesses.

Fireworks hosts:

• Popular open-source models like Mistral 7B. Mixtral and Llama 2 70B.
• FireFunction v1 open-weights function calling model based on Mixtral, 4x faster than GPT-4 with the highest quality for real-world use cases
• Fine-tuning service and hosting - Fine-tune models like Mixtral and serve them serverless at up to 300 token/sec with no extra costs

Getting started with Fireworks.ai

• Sign up at fireworks.ai/login
• Fill this form for getting $50 credits

Learning Resources:

• Getting started with Fireworks.ai
• Models page to get started with models like Mixtral
• Fine-tuning with Fireworks
• Using Firefunction
• Generating Embeddings on Fireworks