Semi-structured RAG - LangChain using Mistral 7B , Qdrant FastEmbed on pdf text with tabular data

If you like to support me financially, It is totally optional and voluntary. Buy me a coffee here: www.buymeacoff...
Many documents contain a mixture of content types, including text and tables.
Semi-structured data can be challenging for conventional RAG for at least two reasons:
• Text splitting may break up tables, corrupting the data in retrieval
• Embedding tables may pose challenges for semantic similarity search
This video shows how to perform RAG on documents with semi-structured data:
• We will use Unstructured to parse both text and tables from documents (PDFs).
• We will use the multi-vector retriever to store raw tables, text along with table summaries better suited for retrieval.
• We will use LCEL to implement the chains used.
We will use Mistral 7B Instruct as our LLM and use Qdrant FastEmbed for our embedding
Colab notebook:
colab.research...
github.com/lan...
huggingface.co...
qdrant.github....
unstructured.io/
Previous video on semi-structured RAG with OpenAI GPT-4: • Semi-structured RAG wi...
If you like such content please subscribe to the channel here:
www.youtube.co...

[@GAAD_Anoop_R]

Could you share the PDF file you have worked on here ?

[@sagarchadha98]

Can you go in detail how extracted text and table looks like? especially table after extracting and before making summaries of table.
Thanks

[@RitheshSreenivasan]

Please debug the colab. Link is shared in the description of the video

[@techthunder4832]

hi sir, can i do this same in amazon sagemaker,or in amazon bedrcok

[@RitheshSreenivasan]

You should be able to do it

[@rnronie38]

Sir, Is this done on paid colab? How can I do this in unpaid colab with cpu? Is it even possible?

[@RitheshSreenivasan]

It should be possible if you use a quantized model. There are other libraries like ollama where you can run it locally on CPU

[@devanshgupta6064]

Table,Text Can we add images data too here?

[@RitheshSreenivasan]

I think if unstructured can handle images, it should work

[@devanshgupta6064]

@@RitheshSreenivasan This video was very informative, could you also try airoboros-13B model someday because it seems to perform better than other open source LLM models or maybe give a shot experimenting with falcon LLM, thanks

[@RitheshSreenivasan]

Yes we can experiment with different models