Update app.py

app.py

@@ -4,18 +4,26 @@ import numpy as np
 from transformers import pipeline, BertTokenizer, BertModel
 import faiss
 import torch
+import json
 import spaces
 # Load CSV data
 data = pd.read_csv('RB10kstats.csv')
 
-# Convert embedding column from string to numpy array
-data['embeddings'] = data['embeddings'].apply(lambda x: np.fromstring(x[1:-1], sep=', '))
+# Convert embedding column from JSON string to numpy array
+data['embeddings'] = data['embeddings'].apply(lambda x: np.array(json.loads(x)))
 
 # Initialize FAISS index
 dimension = len(data['embeddings'][0])
 res = faiss.StandardGpuResources()  # use a single GPU
-index = faiss.IndexFlatL2(dimension)
-gpu_index = faiss.index_cpu_to_gpu(res, 0, index)  # move to GPU
+
+# Check available GPU devices
+num_gpus = faiss.get_num_gpus()
+if num_gpus > 0:
+    gpu_index = faiss.IndexFlatL2(dimension)
+    gpu_index = faiss.index_cpu_to_gpu(res, 0, gpu_index)  # move to GPU
+else:
+    raise RuntimeError("No GPU devices available.")
+
 gpu_index.add(np.stack(data['embeddings'].values))
 
 # Check if GPU is available
@@ -29,7 +37,6 @@ tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
 model = BertModel.from_pretrained('bert-base-uncased').to(device)
 
 # Function to embed the question using BERT
-@spaces.GPU(duration=120)
 def embed_question(question, model, tokenizer):
     inputs = tokenizer(question, return_tensors='pt').to(device)
     with torch.no_grad():
@@ -64,4 +71,4 @@ interface = gr.Interface(
 )
 
 # Launch the Gradio app
-interface.launch()
+interface.launch()