Mistral7b Complete Guide on Colab
HTML-код
- Опубликовано: 30 сен 2023
- The performance of Mistral 7B surpasses that of Llama 2 13B across all criteria and is comparable to Llama 34B. It should be noted that the results reported for Llama 34B are based on the absence of a released Llama 2 34B model. Furthermore, it exhibits significant superiority in terms of code quality and logical analysis benchmarks.
An LLM that is open source provides a level of transparency with respect to its operational mechanisms, architectural design, training data, and methodology, as well as its utilisation. Moreover, the efficient optimisation of an open-source LLM like Mistral-7B has the potential to decrease latency and enhance performance.
We are covering the following technologies in this video:
1. The Lang-Chain framework and integration with Mistral-7B-Instruct.
2. Retrieval Augmented Generation (RAG) with Mistral-7B-Instruct and Chroma DB.
3. GGUF format for commodity hardware (Running on CPU). 
Наука
Astonishing content Man 🔥🔥🔥
Great content and Quality code. Thank you!
Thank you
Thank you son much !
Thanks very much! Two questions: Do I have to install the environnement everytime I use colas? And Can I finetune the Model with books (PDF)?
You can fine tune that the point of it being open source however I'm not sure if you can train the model on PDF's or not yes you can use the content inside the books for sure but not sure if you can feed the data in pdf format or not.
Wow Yogendra! AMazing. I was loking for Mistral and Langchain integration for rAG and chatbot purpose on my tabular data and PDFs. Can this be solved with your tutoria;?
Also what is config file? Is this just the hugging face keys in it?
Hi Anuvrat , Yes above Colab can be used for the use case. Config file is to store Open AI and Hugging Face Tokens . It is not required for specific colab.
Great guide, thank you!! If possible, I'd like to know if is necessary I use "guanaco-llama2-1k" dataset for better RAG, or if it is necessary for finetune, and how. Thanks 🙂
Hello, The purpose here was to use Mistral7B with in-context learning, which means the output received from ChromaDB was directly given to Mistral7B (Part 2 RAG). If your question is if we use Fine-Tune Mistral 7B (meaning for some specific domain, say sales), will it perform better than Raw Mistral7B , short answer is yes. If you can fine-tune your models for specific domains, say sales or healthcare, they will perform better.
how can i host it and get and api to use it apps instead of openai api keys?
Hi Check out huggingface.co/inference-endpoints and together.ai/apis to host your own models if your question is regarding hosting your own models. If your question is hosting the app let us say with an UI there are many solutions like aws.amazon.com/apprunner/
Colab Notebook :: colab.research.google.com/drive/1TWOC6waBUk7SY3T2hQi05skrTaZ7V13W?usp=sharing
Thank you for this! The only one i have found on the internet. Your notebook is dependent on your google drive file path and a json data file. Can you provide this data file as well?
Thank You Mark. You not need any of those. For google Drive you can change it to any path or completely remove (I do it to organize folders). You also do not need json file as my point was all configurations like HuggingFace tokens or Open AI Key I keep in common config file as a best practice. This is common I do irrespective of the problem statement. In above you need not have to load config file at all. So in a Nutshell you can skip these two steps and it will run fine.