Spaces:
Sleeping
Sleeping
Upload chroma_utils.py
Browse files- chroma_utils.py +215 -0
chroma_utils.py
ADDED
|
@@ -0,0 +1,215 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import chromadb
|
| 2 |
+
import os
|
| 3 |
+
import tempfile
|
| 4 |
+
|
| 5 |
+
CHROMA_DIR = "./chromadb_local"
|
| 6 |
+
COLLECTION_NAME = "prompts"
|
| 7 |
+
EMBEDDING_MODEL = "all-MiniLM-L6-v2"
|
| 8 |
+
|
| 9 |
+
import chromadb
|
| 10 |
+
import os
|
| 11 |
+
import tempfile
|
| 12 |
+
import streamlit as st
|
| 13 |
+
|
| 14 |
+
CHROMA_DIR = "./chromadb_persistent" # More descriptive name
|
| 15 |
+
COLLECTION_NAME = "revenue_prompts" # Unique collection name
|
| 16 |
+
EMBEDDING_MODEL = "all-MiniLM-L6-v2"
|
| 17 |
+
|
| 18 |
+
@st.cache_resource
|
| 19 |
+
def get_chroma_client():
|
| 20 |
+
"""Initialize ChromaDB client with proper persistence (cached)"""
|
| 21 |
+
try:
|
| 22 |
+
# Ensure directory exists
|
| 23 |
+
os.makedirs(CHROMA_DIR, exist_ok=True)
|
| 24 |
+
|
| 25 |
+
# Use PersistentClient for true persistence
|
| 26 |
+
client = chromadb.PersistentClient(path=CHROMA_DIR)
|
| 27 |
+
|
| 28 |
+
# Test the client connection
|
| 29 |
+
try:
|
| 30 |
+
client.heartbeat()
|
| 31 |
+
print(f"β
Connected to existing ChromaDB at: {CHROMA_DIR}")
|
| 32 |
+
except:
|
| 33 |
+
print(f"β
Created new ChromaDB at: {CHROMA_DIR}")
|
| 34 |
+
|
| 35 |
+
return client
|
| 36 |
+
except Exception as e:
|
| 37 |
+
print(f"β ChromaDB connection error: {e}")
|
| 38 |
+
st.warning(f"ChromaDB error: {e}")
|
| 39 |
+
# Fallback to in-memory client (will be lost on restart)
|
| 40 |
+
return chromadb.Client()
|
| 41 |
+
|
| 42 |
+
@st.cache_resource
|
| 43 |
+
def get_chroma_collection(_client):
|
| 44 |
+
"""Get or create ChromaDB collection with persistence (cached)"""
|
| 45 |
+
try:
|
| 46 |
+
# Check if collection already exists
|
| 47 |
+
existing_collections = _client.list_collections()
|
| 48 |
+
collection_exists = any(col.name == COLLECTION_NAME for col in existing_collections)
|
| 49 |
+
|
| 50 |
+
if collection_exists:
|
| 51 |
+
collection = _client.get_collection(name=COLLECTION_NAME)
|
| 52 |
+
existing_count = collection.count()
|
| 53 |
+
print(f"β
Using existing collection '{COLLECTION_NAME}' with {existing_count} items")
|
| 54 |
+
st.info(f"π Loaded existing vector database with {existing_count} prompts")
|
| 55 |
+
else:
|
| 56 |
+
collection = _client.create_collection(
|
| 57 |
+
name=COLLECTION_NAME,
|
| 58 |
+
metadata={"description": "Revenue app prompt-SQL pairs"}
|
| 59 |
+
)
|
| 60 |
+
print(f"β
Created new collection '{COLLECTION_NAME}'")
|
| 61 |
+
st.info("π Created new vector database")
|
| 62 |
+
|
| 63 |
+
return collection
|
| 64 |
+
except Exception as e:
|
| 65 |
+
print(f"β Collection error: {e}")
|
| 66 |
+
st.error(f"Collection error: {e}")
|
| 67 |
+
return None
|
| 68 |
+
|
| 69 |
+
@st.cache_resource
|
| 70 |
+
def get_embedding_model():
|
| 71 |
+
"""Get embedding model with caching to avoid reloading"""
|
| 72 |
+
try:
|
| 73 |
+
from sentence_transformers import SentenceTransformer
|
| 74 |
+
import torch
|
| 75 |
+
|
| 76 |
+
device = "cpu" # Use CPU to avoid CUDA issues
|
| 77 |
+
|
| 78 |
+
# Use a persistent cache directory
|
| 79 |
+
cache_dir = os.path.join(CHROMA_DIR, "model_cache")
|
| 80 |
+
os.makedirs(cache_dir, exist_ok=True)
|
| 81 |
+
|
| 82 |
+
print(f"π Loading embedding model: {EMBEDDING_MODEL}")
|
| 83 |
+
model = SentenceTransformer(EMBEDDING_MODEL, device=device, cache_folder=cache_dir)
|
| 84 |
+
print(f"β
Embedding model loaded successfully")
|
| 85 |
+
|
| 86 |
+
return model
|
| 87 |
+
|
| 88 |
+
except Exception as e:
|
| 89 |
+
print(f"β Embedding model error: {e}")
|
| 90 |
+
st.warning(f"Using fallback embedder due to: {e}")
|
| 91 |
+
# Return a simple fallback embedder
|
| 92 |
+
return SimpleEmbedder()
|
| 93 |
+
|
| 94 |
+
def check_database_stats():
|
| 95 |
+
"""Check and display database statistics"""
|
| 96 |
+
try:
|
| 97 |
+
client = get_chroma_client()
|
| 98 |
+
collection = get_chroma_collection(client)
|
| 99 |
+
|
| 100 |
+
if collection:
|
| 101 |
+
count = collection.count()
|
| 102 |
+
return {
|
| 103 |
+
"collection_name": COLLECTION_NAME,
|
| 104 |
+
"total_items": count,
|
| 105 |
+
"database_path": CHROMA_DIR,
|
| 106 |
+
"status": "healthy"
|
| 107 |
+
}
|
| 108 |
+
else:
|
| 109 |
+
return {
|
| 110 |
+
"collection_name": COLLECTION_NAME,
|
| 111 |
+
"total_items": 0,
|
| 112 |
+
"database_path": CHROMA_DIR,
|
| 113 |
+
"status": "error"
|
| 114 |
+
}
|
| 115 |
+
except Exception as e:
|
| 116 |
+
return {
|
| 117 |
+
"collection_name": COLLECTION_NAME,
|
| 118 |
+
"total_items": 0,
|
| 119 |
+
"database_path": CHROMA_DIR,
|
| 120 |
+
"status": f"error: {e}"
|
| 121 |
+
}
|
| 122 |
+
|
| 123 |
+
def initialize_from_jsonl(jsonl_path="prompts.jsonl"):
|
| 124 |
+
"""Initialize ChromaDB from existing JSONL file if database is empty"""
|
| 125 |
+
try:
|
| 126 |
+
if not os.path.exists(jsonl_path):
|
| 127 |
+
return False
|
| 128 |
+
|
| 129 |
+
client = get_chroma_client()
|
| 130 |
+
collection = get_chroma_collection(client)
|
| 131 |
+
embedding_model = get_embedding_model()
|
| 132 |
+
|
| 133 |
+
if not collection or not embedding_model:
|
| 134 |
+
return False
|
| 135 |
+
|
| 136 |
+
# Check if collection is empty
|
| 137 |
+
if collection.count() > 0:
|
| 138 |
+
print(f"β
Database already has {collection.count()} items")
|
| 139 |
+
return True
|
| 140 |
+
|
| 141 |
+
# Load from JSONL
|
| 142 |
+
import json
|
| 143 |
+
prompts_loaded = 0
|
| 144 |
+
|
| 145 |
+
with open(jsonl_path, 'r', encoding='utf-8') as f:
|
| 146 |
+
for i, line in enumerate(f):
|
| 147 |
+
try:
|
| 148 |
+
data = json.loads(line)
|
| 149 |
+
prompt = data.get('prompt', '')
|
| 150 |
+
sql = data.get('sql', '')
|
| 151 |
+
|
| 152 |
+
if prompt and sql:
|
| 153 |
+
# Generate embedding
|
| 154 |
+
embedding = embedding_model.encode([prompt])
|
| 155 |
+
|
| 156 |
+
# Add to collection
|
| 157 |
+
collection.add(
|
| 158 |
+
embeddings=embedding,
|
| 159 |
+
documents=[prompt],
|
| 160 |
+
metadatas=[{"sql": sql, "source": "jsonl_import"}],
|
| 161 |
+
ids=[f"import_{i}"]
|
| 162 |
+
)
|
| 163 |
+
prompts_loaded += 1
|
| 164 |
+
|
| 165 |
+
except Exception as e:
|
| 166 |
+
print(f"β οΈ Error loading line {i}: {e}")
|
| 167 |
+
continue
|
| 168 |
+
|
| 169 |
+
print(f"β
Loaded {prompts_loaded} prompts from {jsonl_path}")
|
| 170 |
+
st.success(f"π Imported {prompts_loaded} existing prompts into vector database")
|
| 171 |
+
return True
|
| 172 |
+
|
| 173 |
+
except Exception as e:
|
| 174 |
+
print(f"β Error initializing from JSONL: {e}")
|
| 175 |
+
return False
|
| 176 |
+
"""Get embedding model with fallback options"""
|
| 177 |
+
try:
|
| 178 |
+
from sentence_transformers import SentenceTransformer
|
| 179 |
+
import torch
|
| 180 |
+
|
| 181 |
+
device = "cpu" # Use CPU to avoid CUDA issues
|
| 182 |
+
cache_dir = tempfile.mkdtemp()
|
| 183 |
+
|
| 184 |
+
# Try to load the model with error handling
|
| 185 |
+
model = SentenceTransformer(EMBEDDING_MODEL, device=device, cache_folder=cache_dir)
|
| 186 |
+
return model
|
| 187 |
+
|
| 188 |
+
except Exception as e:
|
| 189 |
+
print(f"Embedding model error: {e}")
|
| 190 |
+
# Return a simple fallback embedder
|
| 191 |
+
return SimpleEmbedder()
|
| 192 |
+
|
| 193 |
+
class SimpleEmbedder:
|
| 194 |
+
"""Simple fallback embedding function"""
|
| 195 |
+
def encode(self, texts, convert_to_numpy=True):
|
| 196 |
+
import hashlib
|
| 197 |
+
import numpy as np
|
| 198 |
+
|
| 199 |
+
if isinstance(texts, str):
|
| 200 |
+
texts = [texts]
|
| 201 |
+
|
| 202 |
+
embeddings = []
|
| 203 |
+
for text in texts:
|
| 204 |
+
# Create a simple 384-dimensional embedding from text hash
|
| 205 |
+
hash_obj = hashlib.md5(text.encode())
|
| 206 |
+
hash_hex = hash_obj.hexdigest()
|
| 207 |
+
# Convert hex to numbers and pad to 384 dimensions
|
| 208 |
+
embedding = [ord(c) / 255.0 for c in hash_hex]
|
| 209 |
+
# Repeat pattern to get 384 dimensions
|
| 210 |
+
while len(embedding) < 384:
|
| 211 |
+
embedding.extend(embedding[:384-len(embedding)])
|
| 212 |
+
embedding = embedding[:384]
|
| 213 |
+
embeddings.append(embedding)
|
| 214 |
+
|
| 215 |
+
return np.array(embeddings) if convert_to_numpy else embeddings
|