File size: 10,650 Bytes
f13a432 1d31f57 511e86c eecda7f 511e86c eecda7f 511e86c d962317 91cc1cc f13a432 91cc1cc f13a432 d962317 f13a432 9b0b1de 02c0154 9b0b1de 40471fa 9b0b1de 91cc1cc 9b0b1de 02c0154 9b0b1de 00014fa 9b0b1de d5b36ad 9b0b1de 02c0154 9b0b1de 02c0154 91cc1cc f13a432 91cc1cc f13a432 511e86c f13a432 511e86c 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 91cc1cc f13a432 511e86c f13a432 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 |
import streamlit as st
import os
import glob
import torch
import faiss
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModel
from sentence_transformers import CrossEncoder
import pickle
import chromadb
from chromadb.utils import embedding_functions
from huggingface_hub import login
hf_token = os.getenv("HF_TOKEN")
if hf_token:
try:
# Use the official login function
login(token=hf_token)
st.success("Hugging Face token successfully validated (or cached).")
except Exception as e:
# This might reveal a more specific error than AxiosError
st.error(f"Hugging Face login failed (check token validity): {e}")
# Still set the environment variables for downstream libraries
os.environ["HF_TOKEN"] = hf_token
os.environ["HUGGINGFACE_HUB_TOKEN"] = hf_token
else:
st.warning("HF_TOKEN not found. Using anonymous access (may lead to 403 for private models).")
BASE_DIR = "/tmp"
os.makedirs(BASE_DIR, exist_ok=True)
# Global variables
collected_file = f"{BASE_DIR}/collected_data.txt"
vector_db_file = f"{BASE_DIR}/vector_db.faiss"
embedding_file = f"{BASE_DIR}/embeddings.npy"
chunks_file = f"{BASE_DIR}/chunks.pkl"
emb_choice_file = f"{BASE_DIR}/embedding_choice.txt"
index_choice_file = f"{BASE_DIR}/index_choice.txt"
chroma_dir = f"{BASE_DIR}/chroma_db"
os.makedirs(chroma_dir, exist_ok=True)
def bert_encode(model,tokenizer,texts, batch_size=300, device="cpu"):
model.to(device)
all_embeddings = []
with torch.no_grad():
for i in range(0, len(texts), batch_size):
batch_texts = texts[i:i+batch_size]
inputs = tokenizer(batch_texts, return_tensors='pt', padding=True, truncation=True, max_length=512).to(device)
outputs = model(**inputs)
embeddings = outputs.last_hidden_state.mean(dim=1)
all_embeddings.append(embeddings.cpu().numpy().astype("float16")) # save memory
return np.vstack(all_embeddings)
st.title("Retriever Web Application")
st.header("Collect Data -> Chunking & Embedding -> Vector DB Creation -> Inquiry & Reranking")
# Tab 1: Collect Data
tab1, tab2, tab3 = st.tabs(["Collect Data", "DB Formation", "Inquiry Vector DB"])
with tab1:
st.header("Collect Data")
# NEW: Use a text area for data input
text_input = st.text_area(
"Paste your data file content here:",
height=300,
placeholder="Paste your text data here..."
)
collected_file_path = collected_file
# 🌟 FIX: Pass a unique 'key' argument to the button
collect_button_pressed = st.button("Collect", key="collect_data_button_tab1")
if collect_button_pressed and text_input:
all_text = text_input
# Save the content to the temporary file
with open(collected_file_path, "w", encoding="utf-8") as f:
f.write(all_text)
st.success("Collected 1 file's content successfully!")
elif collect_button_pressed and not text_input:
st.warning("Please paste content into the text area before clicking Collect.")
else:
st.write("Waiting for data input...")
# Tab 2: DB Formation
with tab2:
st.header("Vector DB Formation")
chunk_size = st.number_input("Chunk size:", 50, 1000, 200, step=50)
overlap = st.number_input("Overlap size:", 0, 500, 50, step=10)
embedding_choice = st.selectbox("Embedding Technique", ["SentencePiece", "TF-IDF", "BERT"])
index_choice = st.selectbox("Vector DB", ["FAISS","ChromaDB"])
embeddings = None
if st.button("Create DB"):
with open(collected_file, "r", encoding="utf-8") as f:
text_data = f.read()
chunks = [text_data[i:i+chunk_size] for i in range(0, len(text_data), chunk_size-overlap)]
if embedding_choice == "SentencePiece":
model = SentenceTransformer("all-MiniLM-L6-v2",use_auth_token=hf_token)
embeddings = model.encode(chunks, batch_size=300)
elif embedding_choice == "TF-IDF":
vectorizer = TfidfVectorizer()
embeddings = vectorizer.fit_transform(chunks).toarray()
elif embedding_choice == "BERT":
model_name = "bert-base-uncased"
tokenizer = AutoTokenizer.from_pretrained(model_name,token=hf_token)
model = AutoModel.from_pretrained(model_name,token=hf_token)
embeddings = bert_encode(model,tokenizer,chunks)
if index_choice == "FAISS":
dim = len(embeddings[0])
index = faiss.IndexFlatL2(dim)
index.add(np.array(embeddings).astype("float32"))
faiss.write_index(index, vector_db_file)
np.save(embedding_file, embeddings)
else: # ChromaDB
# client = chromadb.PersistentClient(path="chroma_db")
client = chromadb.PersistentClient(path=chroma_dir)
try:
client.delete_collection("rag_collection")
except:
pass
collection = client.get_or_create_collection("rag_collection")
collection.add(
documents=chunks,
embeddings=embeddings,
ids=[str(i) for i in range(len(chunks))]
)
with open(chunks_file, "wb") as f:
pickle.dump(chunks, f)
with open(emb_choice_file, "w") as f:
f.write(embedding_choice)
with open(index_choice_file, "w") as f:
f.write(index_choice)
st.write(f"Saved embeddings with shape: {embeddings.shape}")
# Tab 3: Inquiry Vector DB
with tab3:
st.header("Inquiry Vector DB")
user_query = st.text_area("User Query")
expert_answer = st.text_area("Expert Answer")
k = st.number_input("Number of retrieved data (k):", 1, 20, 5, step=1)
#similarity_metric = st.selectbox("Similarity", ["cosine", "euclidean"])
if st.button("Search"):
# Load chunks and embedding choice and index choice
with open(chunks_file, "rb") as f:
chunks = pickle.load(f)
with open(emb_choice_file, "r") as f:
embedding_choice = f.read().strip()
with open(index_choice_file, "r") as f:
index_choice = f.read().strip()
#display embedding choice and index choice
st.header(f"Using Embedding: {embedding_choice}, Index: {index_choice}")
query_emb = None
if embedding_choice == "SentencePiece":
model = SentenceTransformer("all-MiniLM-L6-v2")
query_emb = model.encode([user_query])
elif embedding_choice == "TF-IDF":
vectorizer = TfidfVectorizer()
vectorizer.fit(chunks) # fit on same chunks
query_emb = vectorizer.transform([user_query]).toarray()
elif embedding_choice == "BERT":
model_name = "bert-base-uncased"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModel.from_pretrained(model_name)
query_emb = bert_encode(model,tokenizer,[user_query])
if index_choice == "ChromaDB":
#display similarity score measure used by chromadb and illustrate what number of score means more similar and its range
st.write("Using ChromaDB with cosine similarity. In cosine similarity, a score closer to 1 means more similarity. " \
"Conversely, a score closer to 0 means less similarity." \
"Cosine similarity scores range from -1 to 1, where 1 indicates perfect similarity, 0 indicates no similarity, and -1 indicates " \
"perfect dissimilarity.")
client = chromadb.PersistentClient(path=chroma_dir)
collection = client.get_or_create_collection("rag_collection")
results = collection.query(
query_embeddings=query_emb.tolist(),
n_results=k,
include=["documents", "distances"]
)
#display retrieved texts and scores beside each other
retrieved_texts = results["documents"][0]
retrieved_scores = results["distances"][0]
st.subheader("Retrieved texts and scores:")
for doc, score in zip(retrieved_texts, retrieved_scores):
st.markdown(f"**Score:** {score:.4f}")
st.write(doc)
st.markdown("---")
else: # FAISS
#display similarity score measure used by FAISS and illustrate what number of score means more similar and its range
st.write("Using FAISS with L2 distance In L2 distance, a smaller score means more similarity." \
"L2 distance scores range from 0 to infinity, where 0 indicates perfect similarity (identical vectors), "
"and larger values indicate less similarity.")
index = faiss.read_index(vector_db_file)
D, I = index.search(query_emb.astype("float32"), k)
#display retrieved texts and scores beside each other
retrieved_texts = [chunks[i] for i in I[0]]
st.subheader("Retrieved texts and scores:")
for doc, score in zip(retrieved_texts, D[0]):
st.markdown(f"**Score:** {score:.4f}")
st.write(doc)
st.markdown("---")
# Reranking
#display similarity score measure used by ReRank and illustrate what number of score means more similar and its range
st.write("Reranking using Cross-ReRank (higher score means more relevance, and lower score means less relevance). " \
"It is relative ranking (higher score = more relevant), not the absolute magnitude.")
reranker = CrossEncoder("cross-encoder/ms-marco-MiniLM-L-6-v2",token=hf_token)
scores = [reranker.predict([(user_query, doc)])[0] for doc in retrieved_texts]
st.subheader("Reranked scores:")
for doc, score in zip(retrieved_texts, scores):
st.markdown(f"**Rerank Score:** {score:.4f}")
st.write(doc)
st.markdown("---")
#meausre relevance if expert answer is provided
if expert_answer.strip():
relevance_scores = [reranker.predict([(expert_answer, doc)])[0] for doc in retrieved_texts]
st.subheader("Relevance to Expert Answer:")
for doc, score in zip(retrieved_texts, relevance_scores):
st.markdown(f"**Relevance Score:** {score:.4f}")
st.write(doc)
st.markdown("---") |