File size: 7,143 Bytes
e67ba51 a337229 e67ba51 a337229 e67ba51 a337229 e67ba51 661eb14 e67ba51 a337229 e67ba51 a337229 661eb14 e67ba51 a337229 e67ba51 a337229 e67ba51 a337229 | 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 | """
In-memory vector store.
Primary: sentence-transformers all-MiniLM-L6-v2 + cosine similarity.
Fallback: BM25 keyword retrieval (if sentence-transformers fails to load).
"""
import hashlib
import math
import re
import numpy as np
import streamlit as st
# ── Embedding model (primary) ─────────────────────────────────────────────────
@st.cache_resource(show_spinner="Loading embedding model…")
def _get_model():
from sentence_transformers import SentenceTransformer
return SentenceTransformer("all-MiniLM-L6-v2")
def _embed(texts: list[str]) -> np.ndarray | None:
try:
model = _get_model()
embs = model.encode(texts, show_progress_bar=False, convert_to_numpy=True)
return embs.astype(np.float32)
except Exception:
return None
# ── BM25 fallback ─────────────────────────────────────────────────────────────
def _tokenize(text: str) -> list[str]:
words = re.findall(r"[a-z0-9]+", text.lower())
bigrams = [f"{words[i]} {words[i+1]}" for i in range(len(words) - 1)]
return words + bigrams
def _bm25_score(q_toks: list[str], doc_toks: list[str],
df: dict[str, int], n_docs: int) -> float:
k1, b, avgdl = 1.5, 0.75, 150.0
dl = len(doc_toks)
tf: dict[str, int] = {}
for t in doc_toks:
tf[t] = tf.get(t, 0) + 1
score = 0.0
for term in set(q_toks):
f = tf.get(term, 0)
if not f:
continue
idf = math.log((n_docs - df.get(term, 0) + 0.5) /
(df.get(term, 0) + 0.5) + 1)
score += idf * f * (k1 + 1) / (f + k1 * (1 - b + b * dl / avgdl))
return score
# ── Collection ────────────────────────────────────────────────────────────────
class _Collection:
def __init__(self, name: str):
self.name = name
self._ids: list[str] = []
self._docs: list[str] = []
self._metas: list[dict] = []
self._embs: np.ndarray = np.empty((0, 384), dtype=np.float32)
# BM25 fallback structures
self._tokens: list[list[str]] = []
self._df: dict[str, int] = {}
def count(self) -> int:
return len(self._docs)
def upsert(self, documents: list[str], ids: list[str],
metadatas: list[dict]) -> None:
new_embs = _embed(documents) # None if model unavailable
for i, (doc, did, meta) in enumerate(zip(documents, ids, metadatas)):
toks = _tokenize(doc)
if did in self._ids:
idx = self._ids.index(did)
for t in set(self._tokens[idx]):
self._df[t] = max(0, self._df.get(t, 1) - 1)
self._docs[idx] = doc
self._metas[idx] = meta
self._tokens[idx] = toks
if new_embs is not None and self._embs.shape[0] > idx:
self._embs[idx] = new_embs[i]
else:
self._ids.append(did)
self._docs.append(doc)
self._metas.append(meta)
self._tokens.append(toks)
if new_embs is not None:
row = new_embs[i:i+1]
self._embs = (np.vstack([self._embs, row])
if self._embs.shape[0] else row.copy())
for t in set(toks):
self._df[t] = self._df.get(t, 0) + 1
def query(self, query_texts: list[str], n_results: int,
where: dict | None = None) -> dict:
empty = {"documents": [[]], "metadatas": [[]], "distances": [[]]}
if not self._docs:
return empty
candidates = [i for i in range(len(self._docs))
if not where or
all(self._metas[i].get(k) == v for k, v in where.items())]
if not candidates:
return empty
# Primary: cosine similarity
if self._embs.shape[0] == len(self._docs):
q_emb = _embed([query_texts[0]])
if q_emb is not None:
q = q_emb[0] / (np.linalg.norm(q_emb[0]) + 1e-9)
cand_embs = self._embs[candidates]
norms = np.linalg.norm(cand_embs, axis=1, keepdims=True) + 1e-9
scores = (cand_embs / norms) @ q
top_idx = sorted(range(len(candidates)),
key=lambda x: scores[x], reverse=True)[:n_results]
top = [candidates[x] for x in top_idx]
return {
"documents": [[self._docs[i] for i in top]],
"metadatas": [[self._metas[i] for i in top]],
"distances": [[float(1 - scores[x]) for x in top_idx]],
}
# Fallback: BM25
q_toks = _tokenize(query_texts[0])
n = len(self._docs)
scored = sorted(candidates,
key=lambda i: _bm25_score(q_toks, self._tokens[i],
self._df, n),
reverse=True)[:n_results]
max_s = _bm25_score(q_toks, self._tokens[scored[0]], self._df, n) if scored else 1.0
return {
"documents": [[self._docs[i] for i in scored]],
"metadatas": [[self._metas[i] for i in scored]],
"distances": [[1 - _bm25_score(q_toks, self._tokens[i],
self._df, n) / max(max_s, 1e-9)
for i in scored]],
}
# ── Public API ────────────────────────────────────────────────────────────────
@st.cache_resource
def _get_store() -> dict[str, _Collection]:
return {}
def get_collection(name: str) -> _Collection:
store = _get_store()
if name not in store:
store[name] = _Collection(name)
return store[name]
def add_chunks(collection: _Collection, chunks: list[dict],
metadatas: list[dict]) -> None:
if not chunks:
return
ids = [hashlib.sha256(c["text"].encode()).hexdigest()[:16] for c in chunks]
collection.upsert(
documents=[c["text"] for c in chunks],
ids=ids,
metadatas=metadatas,
)
def query(collection: _Collection, text: str, k: int = 4,
where: dict | None = None) -> list[dict]:
if collection.count() == 0:
return []
n = min(k, collection.count())
results = collection.query(query_texts=[text], n_results=n, where=where)
return [
{"text": doc, "metadata": meta, "distance": dist}
for doc, meta, dist in zip(
results["documents"][0],
results["metadatas"][0],
results["distances"][0],
)
]
|