Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -3,58 +3,78 @@
|
|
3 |
#api_token = os.getenv("HF_TOKEN").strip()
|
4 |
|
5 |
import torch
|
6 |
-
from
|
7 |
from transformers import AutoModel, AutoTokenizer, BitsAndBytesConfig
|
8 |
-
|
|
|
|
|
9 |
|
|
|
10 |
|
11 |
-
#
|
12 |
bnb_config = BitsAndBytesConfig(
|
13 |
-
load_in_4bit=True,
|
14 |
-
bnb_4bit_quant_type="nf4",
|
15 |
-
bnb_4bit_use_double_quant=True,
|
16 |
-
bnb_4bit_compute_dtype=torch.float16
|
17 |
)
|
18 |
|
19 |
-
# Load
|
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=
|
27 |
)
|
28 |
|
29 |
tokenizer = AutoTokenizer.from_pretrained(
|
30 |
-
"ContactDoctor/Bio-Medical-MultiModal-Llama-3-8B-V1",
|
31 |
trust_remote_code=True
|
32 |
)
|
33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
-
# Define the function to handle the input
|
36 |
-
def process_input(image, question):
|
37 |
-
image = Image.open(image).convert("RGB")
|
38 |
-
msgs = [{'role': 'user', 'content': [image, question]}]
|
39 |
-
res = model.chat(image=image, msgs=msgs, tokenizer=tokenizer, sampling=True, temperature=0.95, stream=True)
|
40 |
-
|
41 |
-
generated_text = ""
|
42 |
-
for new_text in res:
|
43 |
-
generated_text += new_text
|
44 |
-
return generated_text
|
45 |
-
|
46 |
-
# Gradio interface
|
47 |
-
iface = gr.Interface(
|
48 |
-
fn=process_input,
|
49 |
-
inputs=[
|
50 |
-
gr.Image(type="file", label="Upload Image"),
|
51 |
-
gr.Textbox(lines=2, label="Question")
|
52 |
-
],
|
53 |
-
outputs=gr.Textbox(label="Generated Response"),
|
54 |
-
title="BioMedical MultiModal Llama",
|
55 |
-
description="Upload an image and ask a medical question."
|
56 |
-
)
|
57 |
|
58 |
-
if __name__ ==
|
59 |
-
|
60 |
|
|
|
3 |
#api_token = os.getenv("HF_TOKEN").strip()
|
4 |
|
5 |
import torch
|
6 |
+
from flask import Flask, request, jsonify
|
7 |
from transformers import AutoModel, AutoTokenizer, BitsAndBytesConfig
|
8 |
+
from PIL import Image
|
9 |
+
import io
|
10 |
+
import base64
|
11 |
|
12 |
+
app = Flask(__name__)
|
13 |
|
14 |
+
# Quantization configuration
|
15 |
bnb_config = BitsAndBytesConfig(
|
16 |
+
load_in_4bit=True,
|
17 |
+
bnb_4bit_quant_type="nf4",
|
18 |
+
bnb_4bit_use_double_quant=True,
|
19 |
+
bnb_4bit_compute_dtype=torch.float16
|
20 |
)
|
21 |
|
22 |
+
# Load model
|
23 |
model = AutoModel.from_pretrained(
|
24 |
+
"ContactDoctor/Bio-Medical-MultiModal-Llama-3-8B-V1",
|
25 |
+
quantization_config=bnb_config,
|
26 |
+
device_map="auto",
|
27 |
+
torch_dtype=torch.float16,
|
28 |
+
trust_remote_code=True,
|
29 |
+
attn_implementation="flash_attention_2"
|
30 |
)
|
31 |
|
32 |
tokenizer = AutoTokenizer.from_pretrained(
|
33 |
+
"ContactDoctor/Bio-Medical-MultiModal-Llama-3-8B-V1",
|
34 |
trust_remote_code=True
|
35 |
)
|
36 |
|
37 |
+
def decode_base64_image(base64_string):
|
38 |
+
# Decode base64 image
|
39 |
+
image_data = base64.b64decode(base64_string)
|
40 |
+
image = Image.open(io.BytesIO(image_data)).convert('RGB')
|
41 |
+
return image
|
42 |
+
|
43 |
+
@app.route('/analyze', methods=['POST'])
|
44 |
+
def analyze_input():
|
45 |
+
data = request.json
|
46 |
+
question = data.get('question', '')
|
47 |
+
base64_image = data.get('image', None)
|
48 |
+
|
49 |
+
try:
|
50 |
+
# Process with image if provided
|
51 |
+
if base64_image:
|
52 |
+
image = decode_base64_image(base64_image)
|
53 |
+
inputs = model.prepare_inputs_for_generation(
|
54 |
+
input_ids=tokenizer(question, return_tensors="pt").input_ids,
|
55 |
+
images=[image]
|
56 |
+
)
|
57 |
+
outputs = model.generate(**inputs, max_new_tokens=256)
|
58 |
+
else:
|
59 |
+
# Text-only processing
|
60 |
+
inputs = tokenizer(question, return_tensors="pt")
|
61 |
+
outputs = model.generate(**inputs, max_new_tokens=256)
|
62 |
+
|
63 |
+
# Decode response
|
64 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
65 |
+
|
66 |
+
return jsonify({
|
67 |
+
'status': 'success',
|
68 |
+
'response': response
|
69 |
+
})
|
70 |
+
|
71 |
+
except Exception as e:
|
72 |
+
return jsonify({
|
73 |
+
'status': 'error',
|
74 |
+
'message': str(e)
|
75 |
+
}), 500
|
76 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
|
78 |
+
if __name__ == '__main__':
|
79 |
+
app.run(debug=True)
|
80 |
|