Update app.py

app.py

@@ -19,7 +19,7 @@ import json
 cache_dir = '/data/KB'
 os.makedirs(cache_dir, exist_ok=True)
 
-@spaces.GPU(duration=100)
+@spaces.GPU
 def weighted_mean_pooling(hidden, attention_mask):
     attention_mask_ = attention_mask * attention_mask.cumsum(dim=1)
     s = torch.sum(hidden * attention_mask_.unsqueeze(-1).float(), dim=1)
@@ -27,7 +27,7 @@ def weighted_mean_pooling(hidden, attention_mask):
     reps = s / d
     return reps
 
-@spaces.GPU(duration=100)
+@spaces.GPU
 @torch.no_grad()
 def encode(text_or_image_list):
     global model, tokenizer
@@ -63,7 +63,7 @@ def calculate_md5_from_binary(binary_data):
     hash_md5.update(binary_data)
     return hash_md5.hexdigest()
 
-@spaces.GPU(duration=100)
+@spaces.GPU
 def add_pdf_gradio(pdf_file_binary, progress=gr.Progress()):
     global model, tokenizer
     model.eval()
@@ -108,7 +108,7 @@ def add_pdf_gradio(pdf_file_binary, progress=gr.Progress()):
     
     return knowledge_base_name
 
-@spaces.GPU(duration=100)
+@spaces.GPU
 def retrieve_gradio(knowledge_base: str, query: str, topk: int):
     global model, tokenizer
 
@@ -224,7 +224,7 @@ gen_model.to(device)
 print("gen model load success!")
 
 
-@spaces.GPU(duration=50)
+@spaces.GPU
 def answer_question(images, question):
     global gen_model, gen_tokenizer
     # here each element of images is a tuple of (image_path, None).
@@ -257,7 +257,7 @@ with gr.Blocks() as app:
     gr.Markdown("- Currently online demo support PDF document with less than 50 pages due to GPU time limit. Deploy on your own machine for longer PDFs and books.")
     
     with gr.Row():
-        file_input = gr.File(type="binary", label="Step 1: Upload PDF")
+        file_input = gr.File(file_types=["pdf"], file_count="multiple", type="binary", label="Step 1: Upload PDF")
         file_result = gr.Text(label="Knowledge Base ID (remember it, it is re-usable!)")
         process_button = gr.Button("Process PDF (Don't click until PDF upload success)")