diff --git a/config/base.py b/config/base.py
index 8b3b161..ee0e2f9 100644
--- a/config/base.py
+++ b/config/base.py
@@ -8,4 +8,4 @@ GEMMA_LORA_PATH = os.getenv("GEMMA_LORA_PATH")
 BGE_MODEL_PATH = os.getenv("BGE_MODEL_PATH")
 BGE_LORA_PATH = os.getenv("BGE_LORA_PATH")
 
-BATCH_SIZE = 250
\ No newline at end of file
+BATCH_SIZE = 100
\ No newline at end of file
diff --git a/src/serve_embed.py b/src/serve_embed.py
index 6caa30c..b537f5a 100644
--- a/src/serve_embed.py
+++ b/src/serve_embed.py
@@ -55,6 +55,7 @@ def embed_gemma(request: EmbedRequest):
             raise HTTPException(status_code=400, detail="Invalid model")
 
     except Exception as e:
+        torch.cuda.empty_cache()
         raise HTTPException(status_code=500, detail=str(e))
     
     torch.cuda.empty_cache()