improve entities indexing ith GPU

README.md

@@ -214,9 +214,9 @@ Requirements: Docker, nodejs (to build the frontend), and optionally [`uv`](http
 > **Experimental entities indexing**: it can take a lot of time to generate embeddings for entities. So we recommend to run the script to generate embeddings on a machine with GPU (does not need to be a powerful one, but at least with a GPU, checkout [fastembed GPU docs](https://qdrant.github.io/fastembed/examples/FastEmbed_GPU/) to install the GPU drivers and dependencies)
 >
 > ```sh
-> docker compose -f compose.dev.yml up vectordb
+> docker compose -f compose.dev.yml up vectordb -d
 > cd packages/expasy-agent
-> nohup VECTORDB_URL=http://localhost:6334 uv run --extra gpu src/expasy_agent/indexing/index_entities.py --gpu &
+> VECTORDB_URL=http://localhost:6334 nohup uv run --extra gpu src/expasy_agent/indexing/index_entities.py --gpu &
 > ```
 >
 > Then move the entities collection containing the embeddings in `data/qdrant/collections/entities` before starting the stack

compose.yml

@@ -2,7 +2,7 @@ services:
 
   vectordb:
     # https://hub.docker.com/r/qdrant/qdrant/tags
-    image: docker.io/qdrant/qdrant:v1.11.3
+    image: docker.io/qdrant/qdrant:v1.13.2
     # image: qdrant/qdrant:v1.9.2-unprivileged # Unprivileged don't work when mounting a volume
     container_name: vectordb
     restart: unless-stopped

packages/expasy-agent/pyproject.toml

@@ -20,7 +20,8 @@ dependencies = [
     "langgraph >=0.2.61",
     "langchain >=0.2.14",
     # "langchain-community >=0.3.17",
-    "langchain-community @ git+https://github.com/langchain-ai/langchain.git#subdirectory=libs/community",
+    # "langchain-community @ git+https://github.com/langchain-ai/langchain.git#subdirectory=libs/community",
+    "langchain-community @ git+https://github.com/vemonet/langchain.git@fix-fastembed-gpu#subdirectory=libs/community",
     "langchain-openai >=0.1.22",
     "langchain-azure-ai >=0.1.0",
     "langchain-groq >=0.2.4",

packages/expasy-agent/src/expasy_agent/indexing/index_entities.py

@@ -253,7 +253,7 @@ def generate_embeddings_for_entities(gpu: bool = False) -> None:
             retrieve_index_data(entity, docs)
 
     # Uncomment the next line to test with a smaller number of entities
-    docs = docs[:100]
+    # docs = docs[:100]
 
     print(f"Done querying SPARQL endpoints in {(time.time() - start_time) / 60:.2f} minutes, generating embeddings for {len(docs)} entities...")