Update run.sh

run.sh

@@ -1,6 +1,9 @@
 #!/bin/bash
 
 export TRANSFORMERS_CACHE=/root/.cache 
+export CMAKE_ARGS="-DLLAMA_CUBLAS=on"
+export FORCE_CMAKE=1
 
-exec python3 ./ingest.py --device_type cuda
-exec python3 ./run_localGPT.py --device_type cuda
+exec pip install --timeout 100 -r requirements.txt
+exec python ./ingest.py --device_type cuda
+exec python ./run_localGPT.py --device_type cuda