Update constants.py

constants.py

@@ -28,13 +28,13 @@ CHROMA_SETTINGS = Settings(
 )
 
 # Context Window and Max New Tokens
-CONTEXT_WINDOW_SIZE = 2048
+CONTEXT_WINDOW_SIZE = 1024
 MAX_NEW_TOKENS = CONTEXT_WINDOW_SIZE  # int(CONTEXT_WINDOW_SIZE/4)
 
 #### If you get a "not enough space in the buffer" error, you should reduce the values below, start with half of the original values and keep halving the value until the error stops appearing
 
-N_GPU_LAYERS = 20  # Llama-2-70B has 83 layers
-N_BATCH = 512
+N_GPU_LAYERS = 40  # Llama-2-70B has 83 layers
+N_BATCH = 1024
 
 ### From experimenting with the Llama-2-7B-Chat-GGML model on 8GB VRAM, these values work:
 # N_GPU_LAYERS = 20