📌 mistral-haystack collection

Mistral + Haystack Collection: build RAG pipelines that rock 🤘

Collection of notebooks and resources to build Retrieval Augmented Generation pipelines using:

• Mistral models 🤖
• Haystack LLM orchestration framework 🏗️.

💻 For other great Haystack Notebooks, check out the 👩🏻‍🍳 Haystack Cookbook

📓 Notebooks

Model	Haystack version	Link	Details	Author
Mistral-7B-Instruct-v0.1	1.x	🎸 Notebook	RAG on a collection of Rock music resources, using the free Hugging Face Inference API	@anakin87
Mixtral-8x7B-Instruct-v0.1	1.x	📄🚀 Notebook	RAG on a PDF File, using the free Hugging Face Inference API (using the free Hugging Face Inference API)	@AlessandroDiLauro
Mixtral-8x7B-Instruct-v0.1	1.x	🛒 Notebook 📊🔍 Blog post	RAG from CSV, Product description analysis	@AlessandroDiLauro
Mixtral-8x7B-Instruct-v0.1	2.x	🕸️💬 Notebook	RAG on the Web, using the free Hugging Face Inference API	@TuanaCelik
Zephyr-7B Beta	2.x	🪁 Article and notebook	Article on how make this great model (fine-tuned from Mistral) run locally on Colab	@TuanaCelik @anakin87
Mixtral-8x7B-Instruct-v0.1	2.x	🩺💬 Article and notebook	Healthcare chatbot with Mixtral, Haystack, and PubMed	@annthurium
Mixtral-8x7B-Instruct-v0.1	2.x	🇮🇹🇬🇧🎧 Notebook	Multilingual RAG from a podcast	@anakin87
Mixtral-8x7B-Instruct-v0.1	2.x	📰 Notebook	Building a Hacker News Top Stories TL;DR	@TuanaCelik

📚 Resources

• Mixture of Experts Explained

  Great and deep blog post by Hugging Face on the MoE architecture, which is the basis of Mistral 8x7B.

• Zephyr: Direct Distillation of LM Alignment

  Technical report by the Hugging Face H4 team. They explain how they trained Zephyr, a strong 7B model fine-tuned from Mistral.

  The main topic is: ⚗️ how to effectively distill the capabilities of GPT-4 into smaller models? The report is insightful and well worth reading. I have summarized it here.