From a24575102dc9026c995076d4af023c40dbb6793a Mon Sep 17 00:00:00 2001 From: Anand Bodas Date: Wed, 29 May 2024 21:02:50 +0530 Subject: [PATCH] Cleaned up Readme, requirements and VectorDB; Added env variables for HF Token and VectorDB service Signed-off-by: Anand Bodas --- VideoRAGQnA/README.md | 13 +++---- VideoRAGQnA/docs/config.yaml | 14 ++++---- VideoRAGQnA/docs/requirements.txt | 4 ++- .../embedding/generate_store_embeddings.py | 6 ++-- VideoRAGQnA/utils/prompt_handler.py | 4 +-- VideoRAGQnA/video-rag-ui.py | 34 +++++++------------ 6 files changed, 35 insertions(+), 40 deletions(-) diff --git a/VideoRAGQnA/README.md b/VideoRAGQnA/README.md index 0915abc0c..33f99a75b 100644 --- a/VideoRAGQnA/README.md +++ b/VideoRAGQnA/README.md @@ -15,13 +15,13 @@ Visual RAG is a framework that retrives video based on provided user prompt. It ## Prerequisites -There are 10 example videos present in ```files/videos``` along with their description generated by open-source vision model. +There are 10 example videos present in ```video_ingest/videos``` along with their description generated by open-source vision model. If you want these visual RAG to work on your own videos, make sure it matches below format. ## File Structure ```bash -files/ +video_ingest/ . ├── scene_description │ ├── op_10_0320241830.mp4.txt @@ -52,7 +52,8 @@ files/ Install pip requirements ```bash -pip3 install -r VideoRAGQnA/requirements.txt +cd VideoRAGQnA +pip3 install -r docs/requirements.txt ``` The current framework supports both Chroma DB and Intel's VDMS, use either of them, @@ -72,12 +73,12 @@ docker run -d -p 55555:55555 intellabs/vdms:latest Update your choice of db and port in ```config.yaml```. -Generating Image embeddigns and store them into selected db, specify config file location and video input location +Generating Image embeddings and store them into selected db, specify config file location and video input location ```bash -python3 VideoRAGQnA/embedding/generate_store_embeddings.py VideoRAGQnA/docs/config.yaml VideoRAGQnA/video_ingest/videos/ +python3 embedding/generate_store_embeddings.py docs/config.yaml video_ingest/videos/ ``` **Web UI Video RAG** ```bash streamlit run video-rag-ui.py --server.address 0.0.0.0 --server.port 50055 -``` \ No newline at end of file +``` diff --git a/VideoRAGQnA/docs/config.yaml b/VideoRAGQnA/docs/config.yaml index 1f86c967b..391b2f79b 100755 --- a/VideoRAGQnA/docs/config.yaml +++ b/VideoRAGQnA/docs/config.yaml @@ -1,23 +1,23 @@ # Path to all videos -videos: VideoRAGQnA/video_ingest/videos/ +videos: video_ingest/videos/ # Path to video description generated by open-source vision models (ex. video-llama, video-llava, etc.) -description: VideoRAGQnA/video_ingest/scene_description/ +description: video_ingest/scene_description/ # Do you want to extract frames of videos (True if not done already, else False) generate_frames: True # Do you wnat to generate image embeddings? embed_frames: True # Path to store extracted frames -image_output_dir: VideoRAGQnA/video_ingest/frames/ +image_output_dir: video_ingest/frames/ # Path to store metadata files -meta_output_dir: VideoRAGQnA/video_ingest/frame_metadata/ +meta_output_dir: video_ingest/frame_metadata/ # Number of frames to extract per second, # if 24 fps, and this value is 2, then it will extract 12th and 24th frame number_of_frames_per_second: 2 vector_db: choice_of_db: 'vdms' #'chroma' # #Supported databases [vdms, chroma] - host: 10.190.167.193 - port: 55556 #8000 # + host: 0.0.0.0 + port: 55555 #8000 # # LLM path -model_path: VideoRAGQnA/ckpt/llama-2-7b-chat-hf \ No newline at end of file +model_path: meta-llama/Llama-2-7b-chat-hf diff --git a/VideoRAGQnA/docs/requirements.txt b/VideoRAGQnA/docs/requirements.txt index 84a5eb474..d228806b5 100644 --- a/VideoRAGQnA/docs/requirements.txt +++ b/VideoRAGQnA/docs/requirements.txt @@ -6,4 +6,6 @@ streamlit metafunctions sentence-transformers accelerate -vdms \ No newline at end of file +vdms +tzlocal +dateparser diff --git a/VideoRAGQnA/embedding/generate_store_embeddings.py b/VideoRAGQnA/embedding/generate_store_embeddings.py index d8547c7b2..8848fec8a 100644 --- a/VideoRAGQnA/embedding/generate_store_embeddings.py +++ b/VideoRAGQnA/embedding/generate_store_embeddings.py @@ -9,6 +9,7 @@ # Add the parent directory of the current script to the Python path sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))) +VECTORDB_SERVICE_HOST_IP = os.getenv("VECTORDB_SERVICE_HOST_IP", "0.0.0.0") # sys.path.append(os.path.abspath('../utils')) @@ -150,16 +151,17 @@ def retrieval_testing(): meta_output_dir = config['meta_output_dir'] N = config['number_of_frames_per_second'] - host = config['vector_db']['host'] + host = VECTORDB_SERVICE_HOST_IP port = int(config['vector_db']['port']) selected_db = config['vector_db']['choice_of_db'] # Creating DB print ('Creating DB with text and image embedding support, \nIt may take few minutes to download and load all required models if you are running for first time.') + print('Connect to {} at {}:{}'.format(selected_db, host, port)) vs = db.VS(host, port, selected_db) generate_image_embeddings(selected_db) retrieval_testing() - \ No newline at end of file + diff --git a/VideoRAGQnA/utils/prompt_handler.py b/VideoRAGQnA/utils/prompt_handler.py index 500f738b0..35b368fd3 100644 --- a/VideoRAGQnA/utils/prompt_handler.py +++ b/VideoRAGQnA/utils/prompt_handler.py @@ -1,8 +1,8 @@ from jinja2 import Environment, BaseLoader -PROMPT = open("VideoRAGQnA/utils/prompt_template.jinja2").read().strip() +PROMPT = open("utils/prompt_template.jinja2").read().strip() def get_formatted_prompt(scene, prompt): env = Environment(loader=BaseLoader()) template = env.from_string(PROMPT) - return template.render(scene=scene, prompt=prompt) \ No newline at end of file + return template.render(scene=scene, prompt=prompt) diff --git a/VideoRAGQnA/video-rag-ui.py b/VideoRAGQnA/video-rag-ui.py index d9ea6e78d..7c4cc2a55 100644 --- a/VideoRAGQnA/video-rag-ui.py +++ b/VideoRAGQnA/video-rag-ui.py @@ -3,26 +3,25 @@ from embedding.vector_stores import db import time import torch -import streamlit as st import torch import streamlit as st from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer +from transformers import set_seed -from transformers import TextIteratorStreamer from typing import Any, List, Mapping, Optional from langchain_core.callbacks.manager import CallbackManagerForLLMRun from langchain.llms.base import LLM import threading -from transformers import set_seed from utils import config_reader as reader from utils import prompt_handler as ph # from vector_stores import db +HUGGINGFACEHUB_API_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN", "") set_seed(22) if 'config' not in st.session_state.keys(): - st.session_state.config = reader.read_config('VideoRAGQnA/docs/config.yaml') + st.session_state.config = reader.read_config('docs/config.yaml') config = st.session_state.config @@ -51,11 +50,12 @@ @st.cache_resource def load_models(): + #print("HF Token: ", HUGGINGFACEHUB_API_TOKEN) model = AutoModelForCausalLM.from_pretrained( - model_path, torch_dtype=torch.float32, device_map='auto', trust_remote_code=True, + model_path, torch_dtype=torch.float32, device_map='auto', trust_remote_code=True, token=HUGGINGFACEHUB_API_TOKEN ) - tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) + tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True, token=HUGGINGFACEHUB_API_TOKEN) tokenizer.padding_size = 'right' streamer = TextIteratorStreamer(tokenizer, skip_prompt=True) @@ -248,22 +248,12 @@ def display_messages(): 'Find similar videos', 'Man wearing glasses', 'People reading item description', - 'Man wearing khaki pants', - 'Man laughing', - 'Black tshirt guy holding red basket', 'Man holding red shopping basket', - 'Man wearing blue shirt', - 'Man putting object into his pocket', - 'Was there any shoplifting reported?', - 'Was there any shoplifting reported today?', - 'Was there any shoplifting reported in the last 6 hours?', - 'Was there any shoplifting reported last Sunday?', - 'Was there any shoplifting reported last Monday?', - 'Have there been instances of shoplifting?', - 'Have there been instances of shoplifting last Friday?', - 'Have there been any instances of theft or shoplifting in the last 30 minutes?', - 'Have there been any instances of theft or shoplifting in the last 48 hours?', - 'Have there been any instances of theft or shoplifting in the last 72 hours?', + 'Was there any person wearing a blue shirt seen today?', + 'Was there any person wearing a blue shirt seen in the last 6 hours?', + 'Was there any person wearing a blue shirt seen last Sunday?', + 'Was a person wearing glasses seen in the last 30 minutes?', + 'Was a person wearing glasses seen in the last 72 hours?', ), key='example_video' ) @@ -290,4 +280,4 @@ def display_messages(): with col1: display_messages() - handle_message() \ No newline at end of file + handle_message()