Generative AI 101: When to use RAG vs Fine Tuning?
HTML-код
- Опубликовано: 6 сен 2024
- Explore the importance of model performance and comparative analysis of RAG and Fine-tuning strategies.
Our CEO Adit Jain explores the crucial decision-making process of choosing between fine-tuned LLMs, out-of-the-box models, or integrating RAG.
Dive deep into the world of AI applications for personalized insights and enterprise efficiency.
Watch the episode now for a strategic AI roadmap! 💡💬
Standard LLM is a general purpose models which has knowledge of the data it has been trained. This for GPT-3.5 (175B) parameters has been the internet corpus - close to 50B words on the internet in nov 2021. However does not understand your corporate corpus as it not there in the internet and does not understand. RAG is a technique where you would supply this data and use the general purpose knowledge of the LLM to do conversations over. There is no new training that is done, but only new data that is being supplied which otherwise the LLM would not know. Fine Tune LLM is more about the training the LLM itself for performing tasks different from how it would normally. All it takes is modifying atleast one of the parameters (weights and biases) which will alter it parameter.
Please improve sound quality
This!
the echo of the room
RAG is limited by context window of baseline model. Need to keep that in mind
How about using both a fine tuned model and RAG? Wouldn't that be best?
Still not clear pl give with examples we can do knowledge search with fine tuning too no need RAG
i have already completed the fine tuning the llm model from hugging face, with my own data i used it locally, what is the value for this project
I am not hearing this video just because of the sound quality...
Thanks for this information... could you please explain if we have data processing task in that case which which approach is best. i.e. if we needs to process data from the Raw PDF files and needs to store in a organized way which can be easily understandable by the end users.
cool sound quality
What is your linkedin?