æLtorio
commited on
Commit
•
1d6cff4
1
Parent(s):
ca9f0b9
add decriptions
Browse files
app.py
CHANGED
@@ -1,41 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
from transformers import AutoProcessor, Idefics3ForConditionalGeneration, image_utils
|
3 |
import torch
|
|
|
|
|
4 |
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
|
5 |
-
print(f"Using device: {device}")
|
6 |
-
|
7 |
-
# model
|
8 |
-
|
|
|
|
|
|
|
9 |
processor = AutoProcessor.from_pretrained(base_model_path, trust_remote_code=True)
|
|
|
|
|
10 |
model = Idefics3ForConditionalGeneration.from_pretrained(
|
11 |
-
|
12 |
-
|
13 |
|
14 |
-
model
|
|
|
15 |
|
|
|
16 |
def infere(image):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
messages = [
|
18 |
{
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
},
|
24 |
-
{
|
25 |
-
"role": "user",
|
26 |
-
"content": [
|
27 |
-
{"type": "image"},
|
28 |
-
{"type": "text", "text": "What do we see in this image?"},
|
29 |
-
]
|
30 |
-
},
|
31 |
]
|
|
|
|
|
32 |
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
|
|
|
|
|
33 |
inputs = processor(text=prompt, images=[image], return_tensors="pt")
|
34 |
-
|
|
|
35 |
inputs = {k: v.to(device) for k, v in inputs.items()}
|
|
|
|
|
36 |
generated_ids = model.generate(**inputs, max_new_tokens=100)
|
|
|
|
|
37 |
generated_texts = processor.batch_decode(generated_ids, skip_special_tokens=True)
|
|
|
38 |
return generated_texts
|
39 |
|
40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
radiotest.launch(share=True)
|
|
|
1 |
+
# Copyright 2024 Ronan Le Meillat
|
2 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
3 |
+
# you may not use this file except in compliance with the License.
|
4 |
+
# You may obtain a copy of the License at
|
5 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
6 |
+
# Import necessary libraries
|
7 |
import gradio as gr
|
8 |
from transformers import AutoProcessor, Idefics3ForConditionalGeneration, image_utils
|
9 |
import torch
|
10 |
+
|
11 |
+
# Determine the device (GPU or CPU) to run the model on
|
12 |
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
|
13 |
+
print(f"Using device: {device}") # Log the device being used
|
14 |
+
|
15 |
+
# Define the model ID and base model path
|
16 |
+
model_id = "eltorio/IDEFICS3_ROCO"
|
17 |
+
base_model_path = "HuggingFaceM4/Idefics3-8B-Llama3" # or change to local path
|
18 |
+
|
19 |
+
# Initialize the processor from the base model path
|
20 |
processor = AutoProcessor.from_pretrained(base_model_path, trust_remote_code=True)
|
21 |
+
|
22 |
+
# Initialize the model from the base model path and set the torch dtype to bfloat16
|
23 |
model = Idefics3ForConditionalGeneration.from_pretrained(
|
24 |
+
base_model_path, torch_dtype=torch.bfloat16
|
25 |
+
).to(device) # Move the model to the specified device
|
26 |
|
27 |
+
# Load the adapter from the model ID and automatically map it to the device
|
28 |
+
model.load_adapter(model_id, device_map="auto")
|
29 |
|
30 |
+
# Define a function to infer a description from an image
|
31 |
def infere(image):
|
32 |
+
"""
|
33 |
+
Generate a description of a medical image.
|
34 |
+
|
35 |
+
Args:
|
36 |
+
- image (PIL Image): The medical image to describe.
|
37 |
+
|
38 |
+
Returns:
|
39 |
+
- generated_texts (List[str]): A list containing the generated description.
|
40 |
+
"""
|
41 |
+
|
42 |
+
# Define a chat template for the model to respond to
|
43 |
messages = [
|
44 |
{
|
45 |
+
"role": "system",
|
46 |
+
"content": [
|
47 |
+
{"type": "text", "text": "You are a valuable medical doctor and you are looking at an image of your patient."},
|
48 |
+
]
|
49 |
+
},
|
50 |
+
{
|
51 |
+
"role": "user",
|
52 |
+
"content": [
|
53 |
+
{"type": "image"},
|
54 |
+
{"type": "text", "text": "What do we see in this image?"},
|
55 |
+
]
|
56 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
]
|
58 |
+
|
59 |
+
# Apply the chat template and add a generation prompt
|
60 |
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
|
61 |
+
|
62 |
+
# Preprocess the input image and text
|
63 |
inputs = processor(text=prompt, images=[image], return_tensors="pt")
|
64 |
+
|
65 |
+
# Move the inputs to the specified device
|
66 |
inputs = {k: v.to(device) for k, v in inputs.items()}
|
67 |
+
|
68 |
+
# Generate a description with the model
|
69 |
generated_ids = model.generate(**inputs, max_new_tokens=100)
|
70 |
+
|
71 |
+
# Decode the generated IDs into text
|
72 |
generated_texts = processor.batch_decode(generated_ids, skip_special_tokens=True)
|
73 |
+
|
74 |
return generated_texts
|
75 |
|
76 |
+
# Define the title, description, and device description for the Gradio interface
|
77 |
+
title = f"<a href='https://huggingface.co/eltorio/IDEFICS3_ROCO'>IDEFICS3_ROCO</a>: Medical Image to Text <b>running on {device}</b>"
|
78 |
+
desc = "This model generates a description of a medical image."
|
79 |
+
|
80 |
+
device_desc = f"This model is running on {device} 🚀." if device == torch.device('cuda') else f"🐢 This model is running on {device} it will be very (very) slow. If you can donate some GPU time it will be usable 🐢. <a href='https://huggingface.co/eltorio/IDEFICS3_ROCO/discussions'>Please contact us.</a>"
|
81 |
+
|
82 |
+
# Define the long description for the Gradio interface
|
83 |
+
long_desc = f"This model is based on the <a href='https://huggingface.co/eltorio/IDEFICS3_ROCO'>IDEFICS3_ROCO model</a>, which is a multimodal model that can generate text from images. It has been fine-tuned on <a href='https://huggingface.co/datasets/eltorio/ROCO-radiology'>eltorio/ROCO-radiology</a> a dataset of medical images and can generate descriptions of medical images. Try uploading an image of a medical image and see what the model generates!<br><b>{device_desc}</b><br> 2024 - Ronan Le Meillat"
|
84 |
+
|
85 |
+
# Create a Gradio interface with the infere function and specified title and descriptions
|
86 |
+
radiotest = gr.Interface(fn=infere, inputs="image", outputs="text", title=title,
|
87 |
+
description=desc, article=long_desc)
|
88 |
+
|
89 |
+
# Launch the Gradio interface and share it
|
90 |
radiotest.launch(share=True)
|