Spaces:
Sleeping
Sleeping
File size: 7,513 Bytes
7644eac |
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 |
"""
Reranking module for improving retrieval relevance.
Rerankers use more sophisticated models to re-score and re-order initial retrieval
results, significantly improving relevance at the cost of additional computation.
"""
import os
from typing import List, Dict, Any, Optional
from langchain.schema import Document
# Optional imports (graceful degradation if not available)
try:
import cohere
COHERE_AVAILABLE = True
except ImportError:
COHERE_AVAILABLE = False
print("⚠️ Cohere not installed. Install with: pip install cohere")
try:
from sentence_transformers import CrossEncoder
CROSS_ENCODER_AVAILABLE = True
except ImportError:
CROSS_ENCODER_AVAILABLE = False
print("⚠️ sentence-transformers not installed. Install with: pip install sentence-transformers")
class Reranker:
"""
Document reranker using Cohere API or local cross-encoder models.
Reranking is a two-stage retrieval process:
1. Initial retrieval (BM25 + semantic) gets ~20-50 candidates
2. Reranker scores each candidate against the query for final ranking
"""
def __init__(
self,
use_local: bool = False,
cohere_api_key: Optional[str] = None,
cohere_model: str = "rerank-english-v3.0",
local_model: str = "cross-encoder/ms-marco-MiniLM-L-6-v2"
):
"""
Initialize reranker.
Args:
use_local: Use local cross-encoder instead of Cohere API
cohere_api_key: Cohere API key (if using Cohere)
cohere_model: Cohere rerank model name
local_model: Local cross-encoder model name
"""
self.use_local = use_local
self.cohere_client = None
self.cross_encoder = None
self.cohere_model = cohere_model
if use_local:
self._init_local_reranker(local_model)
else:
self._init_cohere_reranker(cohere_api_key)
def _init_cohere_reranker(self, api_key: Optional[str]) -> None:
"""Initialize Cohere reranker."""
if not COHERE_AVAILABLE:
print("❌ Cohere not available. Falling back to local reranker.")
self.use_local = True
self._init_local_reranker()
return
api_key = api_key or os.getenv("COHERE_API_KEY")
if not api_key:
print("❌ COHERE_API_KEY not set. Falling back to local reranker.")
self.use_local = True
self._init_local_reranker()
return
try:
self.cohere_client = cohere.Client(api_key)
print(f"✅ Cohere reranker initialized (model: {self.cohere_model})")
except Exception as e:
print(f"❌ Failed to initialize Cohere: {e}")
print("Falling back to local reranker.")
self.use_local = True
self._init_local_reranker()
def _init_local_reranker(self, model_name: str = "cross-encoder/ms-marco-MiniLM-L-6-v2") -> None:
"""Initialize local cross-encoder reranker."""
if not CROSS_ENCODER_AVAILABLE:
print("❌ sentence-transformers not available. Reranking disabled.")
return
try:
self.cross_encoder = CrossEncoder(model_name)
print(f"✅ Local cross-encoder initialized (model: {model_name})")
except Exception as e:
print(f"❌ Failed to initialize cross-encoder: {e}")
def rerank(
self,
query: str,
documents: List[Document],
top_k: int = 5
) -> List[Dict[str, Any]]:
"""
Rerank documents based on relevance to query.
Args:
query: Search query
documents: List of Document objects to rerank
top_k: Number of top results to return
Returns:
Reranked list of documents with scores
"""
if not documents:
return []
# Use appropriate reranker
if self.use_local and self.cross_encoder:
return self._rerank_local(query, documents, top_k)
elif self.cohere_client:
return self._rerank_cohere(query, documents, top_k)
else:
# No reranker available, return original order
print("⚠️ No reranker available. Returning original order.")
return [
{
'document': doc,
'score': doc.metadata.get('relevance_score', 0.5),
'rank': i + 1
}
for i, doc in enumerate(documents[:top_k])
]
def _rerank_cohere(
self,
query: str,
documents: List[Document],
top_k: int
) -> List[Dict[str, Any]]:
"""Rerank using Cohere API."""
try:
# Prepare documents for Cohere
doc_texts = [doc.page_content for doc in documents]
# Call Cohere rerank API
results = self.cohere_client.rerank(
model=self.cohere_model,
query=query,
documents=doc_texts,
top_n=top_k
)
# Build reranked results
reranked = []
for result in results.results:
reranked.append({
'document': documents[result.index],
'score': result.relevance_score,
'rank': len(reranked) + 1
})
print(f"✅ Cohere reranked {len(documents)} → {len(reranked)} documents")
return reranked
except Exception as e:
print(f"❌ Cohere reranking failed: {e}")
# Fallback to original order
return [
{
'document': doc,
'score': doc.metadata.get('relevance_score', 0.5),
'rank': i + 1
}
for i, doc in enumerate(documents[:top_k])
]
def _rerank_local(
self,
query: str,
documents: List[Document],
top_k: int
) -> List[Dict[str, Any]]:
"""Rerank using local cross-encoder."""
try:
# Prepare query-document pairs
pairs = [[query, doc.page_content] for doc in documents]
# Get relevance scores
scores = self.cross_encoder.predict(pairs)
# Sort by score
scored_docs = list(zip(documents, scores))
scored_docs.sort(key=lambda x: x[1], reverse=True)
# Build results
reranked = []
for doc, score in scored_docs[:top_k]:
reranked.append({
'document': doc,
'score': float(score),
'rank': len(reranked) + 1
})
print(f"✅ Local reranked {len(documents)} → {len(reranked)} documents")
return reranked
except Exception as e:
print(f"❌ Local reranking failed: {e}")
# Fallback to original order
return [
{
'document': doc,
'score': doc.metadata.get('relevance_score', 0.5),
'rank': i + 1
}
for i, doc in enumerate(documents[:top_k])
]
|