qnguyen3 commited on
Commit
45fdec3
·
verified ·
1 Parent(s): 0159119

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +1 -3
app.py CHANGED
@@ -17,8 +17,7 @@ tokenizer = AutoTokenizer.from_pretrained(
17
 
18
  model = LlavaQwen2ForCausalLM.from_pretrained(
19
  'qnguyen3/nanoLLaVA-1.5',
20
- torch_dtype=torch.float16,
21
- attn_implementation="flash_attention_2",
22
  trust_remote_code=True,
23
  device_map='auto')
24
 
@@ -84,7 +83,6 @@ def bot_streaming(message, history):
84
  messages.append({"role": "user", "content": f"<image>\n{message['text']}"})
85
  elif len(history) == 0 and image is None:
86
  messages.append({"role": "user", "content": message['text'] })
87
- model = model.to('cuda')
88
 
89
  # if image is None:
90
  # gr.Error("You need to upload an image for LLaVA to work.")
 
17
 
18
  model = LlavaQwen2ForCausalLM.from_pretrained(
19
  'qnguyen3/nanoLLaVA-1.5',
20
+ torch_dtype=torch.bfloat16,
 
21
  trust_remote_code=True,
22
  device_map='auto')
23
 
 
83
  messages.append({"role": "user", "content": f"<image>\n{message['text']}"})
84
  elif len(history) == 0 and image is None:
85
  messages.append({"role": "user", "content": message['text'] })
 
86
 
87
  # if image is None:
88
  # gr.Error("You need to upload an image for LLaVA to work.")