dfg

app.py

@@ -84,6 +84,9 @@ def model_inference(
     output_text = processor.batch_decode(
         generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
     )
+    del model
+    del processor
+    torch.cuda.empty_cache()
     return output_text[0]
 
 
@@ -124,6 +127,7 @@ def search(query: str, ds, images, k):
         results.append((images[idx])) #, f"Page {idx}"
     del model
     del processor
+    torch.cuda.empty_cache()
     print("done")
     return results
 
@@ -181,6 +185,7 @@ def index_gpu(images, ds):
         ds.extend(list(torch.unbind(embeddings_doc.to("cpu"))))
     del model
     del processor
+    torch.cuda.empty_cache()
     print("done")
     return f"Uploaded and converted {len(images)} pages", ds, images