klbr/khoj
1
0
Fork 0
mirror of https://github.com/khoaliber/khoj.git synced 2026-03-02 21:19:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

Reduce search results for cross-encoder to re-rank to improve search speed

Browse Source 
Search time on my notes reduced from 14s to 4s. Cross-encoder
re-ranking step takes majority time, not the cosine similarity search
This commit is contained in:
Debanjum Singh Solanky
2021-11-17 00:50:28 +05:30
parent 0082631f16
commit 4d37ace3d6
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/search_type/asymmetric.py
View File

@@ -24,7 +24,7 @@ def initialize_model():
"Initialize model for assymetric semantic search. That is, where query smaller than results"
torch.set_num_threads(4)
bi_encoder = SentenceTransformer('sentence-transformers/msmarco-MiniLM-L-6-v3') # The bi-encoder encodes all entries to use for semantic search
top_k = 100 # Number of entries we want to retrieve with the bi-encoder
top_k = 30 # Number of entries we want to retrieve with the bi-encoder
cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2') # The cross-encoder re-ranks the results to improve quality
return bi_encoder, cross_encoder, top_k