Hi everyone, I'm working on the llama-

Table extraction is not good with PDF parsers that parse via text for tables. Sometimes it misses the edge table cases.

You could try using https://llamahub.ai/l/nougat_ocr

It uses OCR to extract information from the table.

Rest for Response generation and embedding you can use open source models. LlamaIndex supports variety of LLMs.

https://docs.llamaindex.ai/en/stable/module_guides/models/llms.html#modules

Thanks for the reply.
Further, I want to work on a Q&A system with a PDF(annual report) where questions can be from the table. still, it should give answers. Can it be possible with llama-index??

Yes

I haven't tried the OCR pdf parser myself but I think somone did and for table purpose.

I have tried sample coding with the txt file as input. Results are too hallucinating. I have used open source model Microsoft phi

Will try today..

If u have any sources pls share

Ah, so llm model with smaller size that too open source are not good at following instructions.

Llamaindex has created a compatibility report for some llm models on various factors:https://docs.llamaindex.ai/en/stable/module_guides/models/llms.html#llm-compatibility-tracking

Okk got it

Basically, I want to try Q&A for PDF which has some paragraphs and complex tables. So here it should be OCR + normal way of RAG right??

Yes

That should work