e9fc99ddc6
Vector search with cross-encoder re-ranking, hybrid BM25+vector retrieval, incremental index updates, and multiple LLM backends (Ollama local, OpenAI API).
90 lines
2.8 KiB
Python
90 lines
2.8 KiB
Python
# query.py
|
||
# Run a querry on a vector store
|
||
# This version implements a CATALOG prompt
|
||
#
|
||
# E.M.F. July 2025
|
||
# August 2025 - updated for nd ssearch
|
||
|
||
from llama_index.core import (
|
||
StorageContext,
|
||
load_index_from_storage,
|
||
ServiceContext,
|
||
Settings,
|
||
)
|
||
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
|
||
from llama_index.llms.ollama import Ollama
|
||
from llama_index.core.postprocessor import SimilarityPostprocessor
|
||
from llama_index.core.prompts import PromptTemplate
|
||
|
||
import logging
|
||
logging.basicConfig(level=logging.DEBUG)
|
||
|
||
|
||
CATALOG_PROMPT = PromptTemplate(
|
||
"""You are a research assistant. You’re given journal snippets (CONTEXT) and a user query.
|
||
Your job is NOT to write an essay but to list the best-matching journal files with a 1–2 sentence rationale.
|
||
|
||
Rules:
|
||
- Use only the CONTEXT; do not invent content.
|
||
- Prefer precise references to passages over generalities.
|
||
- Output exactly:
|
||
1) A brief one-line summary of the overall theme you detect.
|
||
2) A bulleted list: **filename** — brief rationale. If available in the snippet, include date or section hints.
|
||
|
||
CONTEXT:
|
||
{context_str}
|
||
|
||
QUERY: {query_str}
|
||
|
||
Now produce the summary line and the bulleted list of matching files."""
|
||
)
|
||
|
||
# Use a local model to generate
|
||
Settings.llm = Ollama(
|
||
# model="llama3.1:8B", # First model tested
|
||
# model="deepseek-r1:8B", # This model shows its reasoning
|
||
model="gemma3:1b",
|
||
request_timeout=360.0,
|
||
context_window=8000
|
||
)
|
||
|
||
def main():
|
||
# Load embedding model (same as used for vector store)
|
||
embed_model = HuggingFaceEmbedding(model_name="all-mpnet-base-v2")
|
||
Settings.embed_model = embed_model
|
||
|
||
# Load persisted vector store + metadata
|
||
storage_context = StorageContext.from_defaults(persist_dir="./storage")
|
||
index = load_index_from_storage(storage_context)
|
||
|
||
query_engine = index.as_query_engine(
|
||
similarity_top_k=10, # pull wide (tune to taste)
|
||
#response_mode="compact", # concise synthesis
|
||
text_qa_template=CATALOG_PROMPT, # <- custom prompt
|
||
# node_postprocessors=[
|
||
# SimilarityPostprocessor(similarity_cutoff=0.75) # keep strong hits; makes result count flexible
|
||
# ],
|
||
)
|
||
|
||
# Query
|
||
while True:
|
||
q = input("\nEnter your question (or 'exit'): ").strip()
|
||
if q.lower() in ("exit", "quit"):
|
||
break
|
||
print()
|
||
response = query_engine.query(q)
|
||
|
||
# Return the query response and source documents
|
||
print(response.response)
|
||
print("\nSource documents:")
|
||
for sn in response.source_nodes:
|
||
meta = getattr(sn, "metadata", None) or sn.node.metadata
|
||
print(meta.get("file_name"), "---", meta.get("file_path"), getattr(sn, "score", None))
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|
||
|
||
|
||
|
||
|