This code implements one of the multiple ways of multi-model RAG. This project processes a PDF file, retrieves relevant content using Colpali, and generates answers using a multi-modal RAG system. The process includes document indexing, querying, and summarizing with the Gemini model.
Key Components:
- RAGMultiModalModel: Used for document indexing and retrieval.
- PDF Processing: Downloads and processes “Attention is All You Need” paper.
- Gemini Model: Used for content generation from retrieved images and queries.
- Base64 Encoding/Decoding: Manages image data retrieved during search.
Diagram:
Motivation:
To enable efficient querying and content generation from multi-modal documents (PDFs with text and images) in response to natural language queries.
Method Details:
- Indexing: The PDF is indexed using the
RAGMultiModalModel, storing both text and image data. - Querying: Natural language queries retrieve relevant document segments.
- Image Processing: Images from the document are decoded, displayed, and used in conjunction with the Gemini model to generate content.
Benefits:
- Multi-modal support for both text and images.
- Streamlined retrieval and summarization pipeline.
- Flexible content generation using advanced LLMs (Gemini model).
Implementation:
- PDF is indexed, and the content is split into text and image segments.
- A query is run against the indexed document to fetch the relevant results.
- Retrieved image data is decoded and passed through the Gemini model for answer generation.
Summary:
This project integrates document indexing, retrieval, and content generation in a multi-modal setting, enabling efficient queries on complex documents like research papers.
Setup
In [ ]:
!pip install -q byaldi !sudo apt-get install -y poppler-utils # for windows – https://stackoverflow.com/questions/18381713/how-to-install-poppler-on-windows
In [ ]:
!pip install -q git+https://github.com/huggingface/transformers.git qwen-vl-utils flash-attn optimum auto-gptq bitsandbytes
Imports
In [ ]:
import base64 import os os.environ[“HF_token”] = ‘your-huggingface-api-key’ # to download the ColPali model from byaldi import RAGMultiModalModel
In [ ]:
RAG = RAGMultiModalModel.from_pretrained(“vidore/colpali-v1.2”, verbose=1)
Download the “Attention is all you need” paper
In [ ]:
!wget https://arxiv.org/pdf/1706.03762 !mkdir docs !mv 1706.03762 docs/attention_is_all_you_need.pdf
--2024-09-20 12:36:38-- https://arxiv.org/pdf/1706.03762 Resolving arxiv.org (arxiv.org)... 151.101.3.42, 151.101.131.42, 151.101.67.42, ... Connecting to arxiv.org (arxiv.org)|151.101.3.42|:443... connected. HTTP request sent, awaiting response... 200 OK Length: 2215244 (2.1M) [application/pdf] Saving to: ‘1706.03762’ 1706.03762 0%[ ] 0 --.-KB/s 1706.03762 100%[===================>] 2.11M --.-KB/s in 0.01s 2024-09-20 12:36:38 (164 MB/s) - ‘1706.03762’ saved [2215244/2215244
Indexing
In [ ]:
RAG.index( input_path=“./docs/attention_is_all_you_need.pdf”, index_name=“attention_is_all_you_need”, store_collection_with_index=True, # set this to false if you don’t want to store the base64 representation overwrite=True )
Starting from v4.46, the `logits` model output will have the same type as the model (except at train time, where it will always be FP32)
Added page 1 of document 0 to index. Added page 2 of document 0 to index. Added page 3 of document 0 to index. Added page 4 of document 0 to index. Added page 5 of document 0 to index. Added page 6 of document 0 to index. Added page 7 of document 0 to index. Added page 8 of document 0 to index. Added page 9 of document 0 to index. Added page 10 of document 0 to index. Added page 11 of document 0 to index. Added page 12 of document 0 to index. Added page 13 of document 0 to index. Added page 14 of document 0 to index. Added page 15 of document 0 to index. Index exported to .byaldi/attention_is_all_you_need Index exported to .byaldi/attention_is_all_you_need
Out[ ]:
{0: 'docs/attention_is_all_you_need.pdf'}
Query time
In [ ]:
query = “What is the BLEU score of the Transformer (base model)?”
In [ ]:
results = RAG.search(query, k=1)
Actual image data
In [ ]:
image_bytes = base64.b64decode(results[0].base64)
In [ ]:
filename = ‘image.jpg’ # I assume you have a JPG file with open(filename, ‘wb’) as f: f.write(image_bytes)
In [ ]:
from IPython.display import Image display(Image(filename))
Test using gemini-1.5-flash
In [ ]:
import google.generativeai as genai genai.configure(api_key=‘your-api-key’) model = genai.GenerativeModel(model_name=“gemini-1.5-flash”)
In [ ]:
from PIL import Image image = Image.open(filename)
In [ ]:
response = model.generate_content([image, query]) print(response.text)
The BLEU score of the Transformer (base model) is 27.3.
