Skip to content

armgit / LEANN

Public Imported from GitHub
0

[MLsys2026 Best Paper]: https://arxiv.org/abs/2506.08276. RAG on Everything with LEANN. Enjoy 97% storage savings while running a fast, accurate, and 100% private RAG application on your personal device.

upd test/mlx

yichuan520030910320c51d5320fa3f
diff --git a/build_mlx_index.py b/test/build_mlx_index.py
similarity index 74%
rename from build_mlx_index.py
rename to test/build_mlx_index.py
index c7c6e4f..13131ae 100644
--- a/build_mlx_index.py
+++ b/test/build_mlx_index.py
@@ -1,5 +1,5 @@
-from leann.api import LeannBuilder
import os
+from leann.api import LeannBuilder, LeannSearcher, LeannChat
# Define the path for our new MLX-based index
INDEX_PATH = "./mlx_diskann_index/leann"
@@ -10,7 +10,7 @@ else:
print("Initializing LeannBuilder with MLX support...")
# 1. Configure LeannBuilder to use MLX
builder = LeannBuilder(
- backend_name="diskann",
+ backend_name="hnsw",
embedding_model="mlx-community/Qwen3-Embedding-0.6B-4bit-DWQ",
use_mlx=True
)
@@ -32,3 +32,11 @@ else:
builder.build_index(INDEX_PATH)
print("\nSuccessfully built the index with MLX embeddings!")
print(f"Check the metadata file: {INDEX_PATH}.meta.json")
+
+
+ chat = LeannChat(index_path=INDEX_PATH)
+ # add query
+ query = "MLX is an array framework for machine learning on Apple silicon."
+ print(f"Query: {query}")
+ response = chat.ask(query, top_k=3, recompute_beighbor_embeddings=True, complexity=3, beam_width=1)
+ print(f"Response: {response}")