filipealmeida commited on
Commit
18d0ebf
1 Parent(s): 2c40787

Add code to select the number of GPU layers

Browse files
Files changed (1) hide show
  1. app.py +4 -2
app.py CHANGED
@@ -1,6 +1,6 @@
 
1
  import gradio as gr
2
  import logging
3
- import re
4
  from llama_cpp import Llama
5
  from huggingface_hub import hf_hub_download
6
  import sys
@@ -46,6 +46,7 @@ def generate_text(prompt, example):
46
 
47
  parser = argparse.ArgumentParser()
48
  parser.add_argument("--model", help="Path to the model file")
 
49
  args = parser.parse_args()
50
 
51
  if args.model:
@@ -53,7 +54,8 @@ if args.model:
53
  else:
54
  model_path = download_model()
55
 
56
- llm = Llama(model_path=model_path)
 
57
 
58
  # Create a Gradio interface
59
  interface = gr.Interface(
 
1
+ #!/usr/bin/env python3
2
  import gradio as gr
3
  import logging
 
4
  from llama_cpp import Llama
5
  from huggingface_hub import hf_hub_download
6
  import sys
 
46
 
47
  parser = argparse.ArgumentParser()
48
  parser.add_argument("--model", help="Path to the model file")
49
+ parser.add_argument("--gpu", help="How many GPU layers to use", default=0, type=int)
50
  args = parser.parse_args()
51
 
52
  if args.model:
 
54
  else:
55
  model_path = download_model()
56
 
57
+ # TODO n_gpu_layers doesn't seem to be working
58
+ llm = Llama(model_path=model_path, n_gpu_layers=args.gpu)
59
 
60
  # Create a Gradio interface
61
  interface = gr.Interface(