The open source Mixtral 8x7B model launched by Mistral uses the “Expert Hybrid”(MoE) architecture. Different from traditional transformers, the MoE model has multiple built-in expert-level feedforward networks (there are 8 in this model). When reasoning, a gating network is responsible for selecting two experts to work. This setting allows MoE to achieve faster reasoning speeds while maintaining performance comparable to large models. The Mixtral 8x7B model has a total of 46.7B parameters, but only 12.9B parameters are activated during actual reasoning to predict the next Token.
In the “Mistral Getting Started Guide” course, Sophia Yang, Ph.D. of Mistral In the lecture, you will learn:
- Explore Mistral’s open source models (Mistral 7B, Mixstral 8x7B) and business models through API calls and Mistral AI’s Le Chat website.
- Implement the JSON schema to generate structured output that can be directly integrated into large software systems.
- Learn how to use function calls for tool operations, such as using custom Python code to query table data.
- Combine your Big Language Model (LLM) responses with external knowledge sources and use RAG technology to enhance practicality.
- Create a Mistral-driven chat interface that can reference external documents.
This course will help improve your prompt engineering skills.
Explore Mistral’s three open source models (Mistral 7B, Mixstral 8x7B, and the latest Mixstral 8x22B) and the three business models (small, medium and large) that Mistral provides access through Web interfaces and API calls.
Leverage Mistral’s JSON schema to generate LLM responses in a structured JSON format to integrate LLM output into larger software applications.
Use Mistral’s API to call user-defined Python functions to perform tasks such as Web searches or retrieving text from databases, enhancing LLM’s ability to find relevant information to answer user queries.
In this course, you will access Mistral AI’s collection of open source and commercial models, including the Mixstral 8x7B model and the latest Mixstral 8x22B. You will learn how to choose the right model for your use cases and practice features such as effective hinting techniques, function calls, JSON schemas, and Retrieval Enhanced Generation (RAG).
More introductions:
Call tasks via the API to access and prompt the Mistral model and determine whether your task is a simple task (classification), medium task (composing email), or advanced task (coding) complexity, and select the appropriate model considering speed requirements.
Learn to use Mistral’s native function calls, where you can provide LLM tools that can be called on demand to perform tasks that traditional code better performs, such as querying numerical data in a database.
Build a basic RAG system from scratch through similarity search, correctly chunk the data, create embeddings, and implement this tool as a function in a chat system.
Build a chat interface to interact with the Mistral model and ask questions about the documents you upload.
By the end of this course, you will be able to leverage Mistral AI’s leading open source and business models.
Introduction to Mistral is a beginner course for anyone who wants to understand and use Mistral AI’s collection of advanced open source and commercial LLMs. If you have already participated in ChatGPT Prompt Engineering for Developers or Prompt Engineering with Llama 2, then this is a great next step!
If you want to learn more, you can click on the link below the video.
Thank you for watching this video. If you like it, please subscribe and like it. thank
Course link:https://deeplearning.ai/short-courses/getting-started-with-mistral/
Video: