Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -3,15 +3,7 @@ import torch
|
|
3 |
from transformers import AutoModel, AutoTokenizer, BitsAndBytesConfig
|
4 |
import gradio as gr
|
5 |
from PIL import Image
|
6 |
-
|
7 |
-
# First, let's check if flash-attn is installed
|
8 |
-
try:
|
9 |
-
import flash_attn
|
10 |
-
FLASH_ATTN_AVAILABLE = True
|
11 |
-
except ImportError:
|
12 |
-
FLASH_ATTN_AVAILABLE = False
|
13 |
-
print("Flash Attention is not installed. Using default attention mechanism.")
|
14 |
-
print("To install Flash Attention, run: pip install flash-attn --no-build-isolation")
|
15 |
|
16 |
# Get API token from environment variable
|
17 |
api_token = os.getenv("HF_TOKEN").strip()
|
@@ -24,23 +16,15 @@ bnb_config = BitsAndBytesConfig(
|
|
24 |
bnb_4bit_compute_dtype=torch.float16
|
25 |
)
|
26 |
|
27 |
-
# Initialize model with conditional Flash Attention
|
28 |
-
model_args = {
|
29 |
-
"quantization_config": bnb_config,
|
30 |
-
"device_map": "auto",
|
31 |
-
"torch_dtype": torch.float16,
|
32 |
-
"trust_remote_code": True,
|
33 |
-
"token": api_token
|
34 |
-
}
|
35 |
-
|
36 |
-
# Only add flash attention if available
|
37 |
-
if FLASH_ATTN_AVAILABLE:
|
38 |
-
model_args["attn_implementation"] = "flash_attention_2"
|
39 |
-
|
40 |
# Initialize model and tokenizer
|
41 |
model = AutoModel.from_pretrained(
|
42 |
"ContactDoctor/Bio-Medical-MultiModal-Llama-3-8B-V1",
|
43 |
-
|
|
|
|
|
|
|
|
|
|
|
44 |
)
|
45 |
|
46 |
tokenizer = AutoTokenizer.from_pretrained(
|
@@ -100,11 +84,6 @@ demo = gr.Interface(
|
|
100 |
|
101 |
# Launch the Gradio app
|
102 |
if __name__ == "__main__":
|
103 |
-
# Print installation instructions if Flash Attention is not available
|
104 |
-
if not FLASH_ATTN_AVAILABLE:
|
105 |
-
print("\nTo enable Flash Attention 2 for better performance, please install it using:")
|
106 |
-
print("pip install flash-attn --no-build-isolation")
|
107 |
-
|
108 |
demo.launch(
|
109 |
share=True,
|
110 |
server_name="0.0.0.0",
|
|
|
3 |
from transformers import AutoModel, AutoTokenizer, BitsAndBytesConfig
|
4 |
import gradio as gr
|
5 |
from PIL import Image
|
6 |
+
from torchvision.transforms import ToTensor
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
|
8 |
# Get API token from environment variable
|
9 |
api_token = os.getenv("HF_TOKEN").strip()
|
|
|
16 |
bnb_4bit_compute_dtype=torch.float16
|
17 |
)
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
# Initialize model and tokenizer
|
20 |
model = AutoModel.from_pretrained(
|
21 |
"ContactDoctor/Bio-Medical-MultiModal-Llama-3-8B-V1",
|
22 |
+
quantization_config=bnb_config,
|
23 |
+
device_map="auto",
|
24 |
+
torch_dtype=torch.float16,
|
25 |
+
trust_remote_code=True,
|
26 |
+
attn_implementation="flash_attention_2",
|
27 |
+
token=api_token
|
28 |
)
|
29 |
|
30 |
tokenizer = AutoTokenizer.from_pretrained(
|
|
|
84 |
|
85 |
# Launch the Gradio app
|
86 |
if __name__ == "__main__":
|
|
|
|
|
|
|
|
|
|
|
87 |
demo.launch(
|
88 |
share=True,
|
89 |
server_name="0.0.0.0",
|