Fix.

app.py

@@ -13,11 +13,9 @@ def get_rag_qa() -> dict:
     torch.cuda.empty_cache()
     return load_all(
         embedder_path="Snowflake/snowflake-arctic-embed-l",
-        embedder_device="cpu",
         context_file="data/bioasq_contexts.jsonl",
         index_file="data/bioasq_contexts__snowflake-arctic-embed-l__float32_hnsw.index",
         reader_path="meta-llama/Llama-3.2-1B-Instruct",
-        reader_device="mps",
     )
 
 

rag.py

@@ -92,15 +92,14 @@ def construct_prompt(contexts: list[str], question: str) -> list[dict]:
 
 def load_all(
     embedder_path: str,
-    embedder_device: str,
     context_file: str,
     index_file: str,
     reader_path: str,
-    reader_device: str,
 ) -> tuple[SentenceTransformer, list[str], faiss.Index, TextGenerationPipeline]:
-    embedder = load_embedder(embedder_path, embedder_device)
+    embedder = load_embedder(embedder_path, "cpu")
     contexts = load_contexts(context_file)
     index = load_index(index_file)
+    reader_device = "cuda" if torch.cuda.is_available() else "cpu"
     reader = load_reader(reader_path, reader_device)
 
     return {

requirements.txt

@@ -68,7 +68,6 @@ setuptools==75.6.0
 six==1.17.0
 smmap==5.0.1
 stack-data==0.6.3
-streamlit==1.41.1
 sympy==1.13.1
 tenacity==9.0.0
 threadpoolctl==3.5.0