Spaces:
Sleeping
Sleeping
app.py
CHANGED
@@ -4,11 +4,9 @@ import torch
|
|
4 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
5 |
import gradio as gr
|
6 |
import matplotlib.pyplot as plt
|
7 |
-
from datasets import load_dataset
|
8 |
-
from evaluate import load # For evaluation metrics
|
9 |
|
10 |
# Model setup
|
11 |
-
device = torch.device('
|
12 |
dtype = torch.float32
|
13 |
model_name_or_path = 'GoodBaiBai88/M3D-LaMed-Phi-3-4B'
|
14 |
proj_out_num = 256
|
@@ -16,8 +14,8 @@ proj_out_num = 256
|
|
16 |
# Load model and tokenizer
|
17 |
model = AutoModelForCausalLM.from_pretrained(
|
18 |
model_name_or_path,
|
19 |
-
torch_dtype=
|
20 |
-
device_map=
|
21 |
trust_remote_code=True
|
22 |
)
|
23 |
|
@@ -29,50 +27,43 @@ tokenizer = AutoTokenizer.from_pretrained(
|
|
29 |
trust_remote_code=True
|
30 |
)
|
31 |
|
32 |
-
# Load the M3D-Cap dataset
|
33 |
-
dataset = load_dataset("GoodBaiBai88/M3D-Cap")
|
34 |
-
|
35 |
# Chat history storage
|
36 |
chat_history = []
|
37 |
current_image = None
|
38 |
|
39 |
def extract_and_display_images(image_path):
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
except Exception as e:
|
57 |
-
return f"Error processing image: {str(e)}"
|
58 |
|
59 |
def process_image(question):
|
60 |
global current_image
|
61 |
if current_image is None:
|
62 |
return "Please upload an image first."
|
63 |
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
except Exception as e:
|
75 |
-
return f"Error generating response: {str(e)}"
|
76 |
|
77 |
def chat_interface(question):
|
78 |
global chat_history
|
@@ -80,48 +71,13 @@ def chat_interface(question):
|
|
80 |
chat_history.append((question, response))
|
81 |
return chat_history
|
82 |
|
|
|
83 |
def upload_image(image):
|
84 |
global current_image
|
85 |
current_image = image.name
|
86 |
extracted_image_path = extract_and_display_images(current_image)
|
87 |
return "Image uploaded and processed successfully!", extracted_image_path
|
88 |
|
89 |
-
def test_model_with_dataset():
|
90 |
-
# Load evaluation metrics
|
91 |
-
bleu = load("bleu")
|
92 |
-
rouge = load("rouge")
|
93 |
-
|
94 |
-
# Initialize lists to store predictions and references
|
95 |
-
predictions = []
|
96 |
-
references = []
|
97 |
-
|
98 |
-
# Iterate over the dataset
|
99 |
-
for example in dataset['train']: # Use 'train', 'validation', or 'test' split
|
100 |
-
image_path = example['image'] # Assuming 'image' contains the path to the .npy file
|
101 |
-
question = example['caption'] # Assuming 'caption' contains the question or caption
|
102 |
-
|
103 |
-
# Upload the image
|
104 |
-
upload_image({"name": image_path})
|
105 |
-
|
106 |
-
# Get the model's response
|
107 |
-
response = process_image(question)
|
108 |
-
|
109 |
-
# Store predictions and references
|
110 |
-
predictions.append(response)
|
111 |
-
references.append(question)
|
112 |
-
|
113 |
-
# Print results for debugging
|
114 |
-
print(f"Question: {question}")
|
115 |
-
print(f"Model Response: {response}")
|
116 |
-
print("---")
|
117 |
-
|
118 |
-
# Compute evaluation metrics
|
119 |
-
bleu_score = bleu.compute(predictions=predictions, references=references)
|
120 |
-
rouge_score = rouge.compute(predictions=predictions, references=references)
|
121 |
-
|
122 |
-
print(f"BLEU Score: {bleu_score}")
|
123 |
-
print(f"ROUGE Score: {rouge_score}")
|
124 |
-
|
125 |
# Gradio UI
|
126 |
with gr.Blocks(theme=gr.themes.Soft()) as chat_ui:
|
127 |
gr.Markdown("ICliniq AI-Powered Medical Image Analysis Workspace")
|
@@ -139,7 +95,4 @@ with gr.Blocks(theme=gr.themes.Soft()) as chat_ui:
|
|
139 |
submit_button.click(chat_interface, question_input, chat_list)
|
140 |
question_input.submit(chat_interface, question_input, chat_list)
|
141 |
|
142 |
-
# Uncomment to test the model with the dataset
|
143 |
-
# test_model_with_dataset()
|
144 |
-
|
145 |
chat_ui.launch()
|
|
|
4 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
5 |
import gradio as gr
|
6 |
import matplotlib.pyplot as plt
|
|
|
|
|
7 |
|
8 |
# Model setup
|
9 |
+
device = torch.device('cpu') # Use 'cuda' if GPU is available
|
10 |
dtype = torch.float32
|
11 |
model_name_or_path = 'GoodBaiBai88/M3D-LaMed-Phi-3-4B'
|
12 |
proj_out_num = 256
|
|
|
14 |
# Load model and tokenizer
|
15 |
model = AutoModelForCausalLM.from_pretrained(
|
16 |
model_name_or_path,
|
17 |
+
torch_dtype=torch.float32,
|
18 |
+
device_map='cpu',
|
19 |
trust_remote_code=True
|
20 |
)
|
21 |
|
|
|
27 |
trust_remote_code=True
|
28 |
)
|
29 |
|
|
|
|
|
|
|
30 |
# Chat history storage
|
31 |
chat_history = []
|
32 |
current_image = None
|
33 |
|
34 |
def extract_and_display_images(image_path):
|
35 |
+
npy_data = np.load(image_path)
|
36 |
+
if npy_data.ndim == 4 and npy_data.shape[1] == 32:
|
37 |
+
npy_data = npy_data[0]
|
38 |
+
elif npy_data.ndim != 3 or npy_data.shape[0] != 32:
|
39 |
+
return "Invalid .npy file format. Expected shape (1, 32, 256, 256) or (32, 256, 256)."
|
40 |
+
|
41 |
+
fig, axes = plt.subplots(4, 8, figsize=(12, 6))
|
42 |
+
for i, ax in enumerate(axes.flat):
|
43 |
+
ax.imshow(npy_data[i], cmap='gray')
|
44 |
+
ax.axis('off')
|
45 |
+
|
46 |
+
image_output = "extracted_images.png"
|
47 |
+
plt.savefig(image_output, bbox_inches='tight')
|
48 |
+
plt.close()
|
49 |
+
return image_output
|
50 |
+
|
|
|
|
|
51 |
|
52 |
def process_image(question):
|
53 |
global current_image
|
54 |
if current_image is None:
|
55 |
return "Please upload an image first."
|
56 |
|
57 |
+
image_np = np.load(current_image)
|
58 |
+
image_tokens = "<im_patch>" * proj_out_num
|
59 |
+
input_txt = image_tokens + question
|
60 |
+
input_id = tokenizer(input_txt, return_tensors="pt")['input_ids'].to(device=device)
|
61 |
+
|
62 |
+
image_pt = torch.from_numpy(image_np).unsqueeze(0).to(dtype=dtype, device=device)
|
63 |
+
generation = model.generate(image_pt, input_id, max_new_tokens=256, do_sample=True, top_p=0.9, temperature=1.0)
|
64 |
+
generated_texts = tokenizer.batch_decode(generation, skip_special_tokens=True)
|
65 |
+
return generated_texts[0]
|
66 |
+
|
|
|
|
|
67 |
|
68 |
def chat_interface(question):
|
69 |
global chat_history
|
|
|
71 |
chat_history.append((question, response))
|
72 |
return chat_history
|
73 |
|
74 |
+
|
75 |
def upload_image(image):
|
76 |
global current_image
|
77 |
current_image = image.name
|
78 |
extracted_image_path = extract_and_display_images(current_image)
|
79 |
return "Image uploaded and processed successfully!", extracted_image_path
|
80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
# Gradio UI
|
82 |
with gr.Blocks(theme=gr.themes.Soft()) as chat_ui:
|
83 |
gr.Markdown("ICliniq AI-Powered Medical Image Analysis Workspace")
|
|
|
95 |
submit_button.click(chat_interface, question_input, chat_list)
|
96 |
question_input.submit(chat_interface, question_input, chat_list)
|
97 |
|
|
|
|
|
|
|
98 |
chat_ui.launch()
|