Spaces:
Sleeping
Sleeping
File size: 4,938 Bytes
54656fc | 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 | """
Build or refresh a Pinecone namespace for retrieval evaluation.
Default behavior:
- index qa_dataset_generation/data/test_notices_2025.json
For full 2025/2026 evaluation, pass --full. Use --force to re-embed notices even
when their manifest entries look unchanged.
"""
from __future__ import annotations
import argparse
import json
import logging
import os
import sys
from pathlib import Path
ROOT = Path(__file__).parent.parent
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
DEFAULT_CORPORA = [ROOT / "qa_dataset_generation" / "data" / "test_notices_2025.json"]
FULL_EVAL_CORPORA = [
ROOT / "qa_dataset_generation" / "data" / "test_notices_2025.json",
ROOT / "data" / "data_2026.json",
]
DEFAULT_EVAL_NAMESPACE = "eval"
DEFAULT_EVAL_STATE = ROOT / "pinecone_index" / "eval"
def load_notices(paths: list[Path]) -> list[dict]:
seen_urls = set()
notices = []
for path in paths:
with path.open(encoding="utf-8") as f:
data = json.load(f)
for item in data:
url = item.get("url")
if not url or url in seen_urls:
continue
seen_urls.add(url)
notices.append(
{
"title": item.get("title", ""),
"url": url,
"date": item.get("date", ""),
"body": item.get("body", ""),
"category": item.get("category", ""),
}
)
return notices
def main() -> None:
parser = argparse.ArgumentParser(description="Build a Pinecone evaluation namespace.")
parser.add_argument(
"--corpus",
type=Path,
action="append",
default=None,
help="Corpus JSON file. Can be passed multiple times.",
)
parser.add_argument(
"--full",
action="store_true",
help="Index both the 2025 QA corpus and current 2026 notices.",
)
parser.add_argument(
"--namespace",
default=DEFAULT_EVAL_NAMESPACE,
help="Pinecone namespace to index.",
)
parser.add_argument(
"--state-dir",
type=Path,
default=DEFAULT_EVAL_STATE,
help="Local manifest/cache directory for the evaluation namespace.",
)
parser.add_argument(
"--force",
action="store_true",
help="Re-embed notices even when manifest entries look unchanged.",
)
parser.add_argument(
"--sync-deletions",
action="store_true",
help="Delete vectors for notices no longer present in the selected corpora.",
)
parser.add_argument(
"--pooling",
choices=["cls", "mean"],
default="cls",
help="Pooling mode used by SimCSEEmbedder.",
)
parser.add_argument(
"--model",
default=None,
help="Embedding model name/path. Defaults to api.core.config.BASE_MODEL_EMBED.",
)
parser.add_argument(
"--backend",
choices=["simcse", "sentence-transformers"],
default="simcse",
help="Embedding backend.",
)
parser.add_argument("--notice-batch-size", type=int, default=20)
parser.add_argument("--embed-batch-size", type=int, default=16)
args = parser.parse_args()
default_corpora = FULL_EVAL_CORPORA if args.full else DEFAULT_CORPORA
corpora = [path.resolve() for path in (args.corpus or default_corpora)]
state_dir = args.state_dir.resolve()
os.environ["VECTOR_DB"] = "pinecone"
os.environ["PINECONE_NAMESPACE"] = args.namespace
os.environ["VECTOR_STATE_PATH"] = str(state_dir)
os.environ["PINECONE_CACHE_PATH"] = str(state_dir / "pinecone_chunks.json")
os.environ["SIMCSE_POOLING"] = args.pooling
os.environ["EMBEDDER_BACKEND"] = args.backend
if args.model:
os.environ["BASE_MODEL_EMBED"] = args.model
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
from api.core.models import get_vector_collection, index_notices
notices = load_notices(corpora)
if not notices:
raise SystemExit("No notices loaded.")
print("=" * 72)
print("Building Pinecone evaluation namespace")
print("=" * 72)
print(f"Namespace : {args.namespace}")
print(f"State dir : {state_dir}")
print(f"Backend : {args.backend}")
print(f"Model : {args.model or '(config default)'}")
print(f"Pooling : {args.pooling}")
print("Corpora:")
for corpus in corpora:
print(f" - {corpus}")
print(f"Notices : {len(notices)} unique URLs")
indexed = index_notices(
notices,
force=args.force,
sync_deletions=args.sync_deletions,
notice_batch_size=args.notice_batch_size,
embed_batch_size=args.embed_batch_size,
)
print(f"Indexed notices : {indexed}")
print(f"Pinecone vectors: {get_vector_collection().count():,}")
if __name__ == "__main__":
main()
|