Improve LLAMA ChromaDB Query Performance
Budget: $30 – $250 USD
Hello,
I am currently experiencing a critical issue with a Python script integrating LLAMA 2, Chroma
DB, and LangChain for document retrieval and processing. This system was operational previously
but has recently started malfunctioning, particularly in retrieving accurate information from the
VectorDB.
Project Overview: The script monitors a specific directory for new PDF files. Upon detecting a
new file, it performs OCR (Optical Character Recognition) using EasyOCR to extract text, which is
then stored in a ChromaDB. Subsequently, it should allow querying LLAMA 2 to search for the
most relevant document and provide answers to questions. However, instead of retrieving accurate
information, the system now generates incorrect or "fake" information.
Specific Issues:
•••••The retrieval process from VectorDB is not functioning as expected.
Despite correct document identification, LLAMA 2 generates inaccurate responses.
The script currently does not support processing multiple questions simultaneously, a
feature crucial for our operation.
The EasyOCR component requires a queue system to efficiently handle the volume of PDFs
produced by a scanner, which scans hundreds of pages per minute.
The environment requires setup for MySQL for file management and adjustment in the .env
file for MySQL and the model path (currently using GGUF models).
I am currently experiencing a critical issue with a Python script integrating LLAMA 2, Chroma
DB, and LangChain for document retrieval and processing. This system was operational previously
but has recently started malfunctioning, particularly in retrieving accurate information from the
VectorDB.
Project Overview: The script monitors a specific directory for new PDF files. Upon detecting a
new file, it performs OCR (Optical Character Recognition) using EasyOCR to extract text, which is
then stored in a ChromaDB. Subsequently, it should allow querying LLAMA 2 to search for the
most relevant document and provide answers to questions. However, instead of retrieving accurate
information, the system now generates incorrect or "fake" information.
Specific Issues:
•••••The retrieval process from VectorDB is not functioning as expected.
Despite correct document identification, LLAMA 2 generates inaccurate responses.
The script currently does not support processing multiple questions simultaneously, a
feature crucial for our operation.
The EasyOCR component requires a queue system to efficiently handle the volume of PDFs
produced by a scanner, which scans hundreds of pages per minute.
The environment requires setup for MySQL for file management and adjustment in the .env
file for MySQL and the model path (currently using GGUF models).