Spaces:
Running
on
Zero
Running
on
Zero
Upload 3 files
Browse files- src/paligemma/__init__.py +0 -0
- src/paligemma/model.py +53 -0
- src/paligemma/response.py +71 -0
src/paligemma/__init__.py
ADDED
File without changes
|
src/paligemma/model.py
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Necessary imports
|
2 |
+
import os
|
3 |
+
import sys
|
4 |
+
from dotenv import load_dotenv
|
5 |
+
from typing import Any
|
6 |
+
import torch
|
7 |
+
from transformers import PaliGemmaForConditionalGeneration, PaliGemmaProcessor
|
8 |
+
|
9 |
+
# Local imports
|
10 |
+
from src.logger import logging
|
11 |
+
from src.exception import CustomExceptionHandling
|
12 |
+
|
13 |
+
|
14 |
+
# Load the Environment Variables from .env file
|
15 |
+
load_dotenv()
|
16 |
+
|
17 |
+
# Access token for using the model
|
18 |
+
access_token = os.environ.get("ACCESS_TOKEN")
|
19 |
+
|
20 |
+
|
21 |
+
def load_model_and_processor(model_name: str, device: str) -> Any:
|
22 |
+
"""
|
23 |
+
Load the model and processor.
|
24 |
+
|
25 |
+
Args:
|
26 |
+
- model_name (str): The name of the model to load.
|
27 |
+
- device (str): The device to load the model onto.
|
28 |
+
|
29 |
+
Returns:
|
30 |
+
- model: The loaded model.
|
31 |
+
- processor: The loaded processor.
|
32 |
+
"""
|
33 |
+
try:
|
34 |
+
# Load the model and processor
|
35 |
+
model = (
|
36 |
+
PaliGemmaForConditionalGeneration.from_pretrained(
|
37 |
+
model_name, torch_dtype=torch.bfloat16, token=access_token
|
38 |
+
)
|
39 |
+
.eval()
|
40 |
+
.to(device)
|
41 |
+
)
|
42 |
+
processor = PaliGemmaProcessor.from_pretrained(model_name, token=access_token)
|
43 |
+
|
44 |
+
# Log the successful loading of the model and processor
|
45 |
+
logging.info("Model and processor loaded successfully.")
|
46 |
+
|
47 |
+
# Return the model and processor
|
48 |
+
return model, processor
|
49 |
+
|
50 |
+
# Handle exceptions that may occur during model and processor loading
|
51 |
+
except Exception as e:
|
52 |
+
# Custom exception handling
|
53 |
+
raise CustomExceptionHandling(e, sys) from e
|
src/paligemma/response.py
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Necessary imports
|
2 |
+
import sys
|
3 |
+
import PIL.Image
|
4 |
+
import torch
|
5 |
+
import gradio as gr
|
6 |
+
import spaces
|
7 |
+
|
8 |
+
# Local imports
|
9 |
+
from src.config import device, model_name
|
10 |
+
from src.paligemma.model import load_model_and_processor
|
11 |
+
from src.logger import logging
|
12 |
+
from src.exception import CustomExceptionHandling
|
13 |
+
|
14 |
+
|
15 |
+
# Language dictionary
|
16 |
+
language_dict = {
|
17 |
+
"English": "en",
|
18 |
+
"Spanish": "es",
|
19 |
+
"French": "fr",
|
20 |
+
}
|
21 |
+
|
22 |
+
# Model and processor
|
23 |
+
model, processor = load_model_and_processor(model_name, device)
|
24 |
+
|
25 |
+
|
26 |
+
@spaces.GPU
|
27 |
+
def caption_image(image: PIL.Image.Image, max_new_tokens: int, language: str) -> str:
|
28 |
+
"""
|
29 |
+
Generates a caption based on the given image using the model.
|
30 |
+
|
31 |
+
Args:
|
32 |
+
- image (PIL.Image.Image): The input image to be processed.
|
33 |
+
- max_new_tokens (int): The maximum number of new tokens to generate.
|
34 |
+
- language (str): The language of the generated caption.
|
35 |
+
|
36 |
+
Returns:
|
37 |
+
str: The generated caption text.
|
38 |
+
"""
|
39 |
+
try:
|
40 |
+
# Check if image is None
|
41 |
+
if not image:
|
42 |
+
gr.Warning("Please provide an image.")
|
43 |
+
|
44 |
+
# Prepare the inputs
|
45 |
+
language = language_dict[language]
|
46 |
+
prompt = f"<image>caption {language}"
|
47 |
+
model_inputs = (
|
48 |
+
processor(text=prompt, images=image, return_tensors="pt")
|
49 |
+
.to(torch.bfloat16)
|
50 |
+
.to(device)
|
51 |
+
)
|
52 |
+
input_len = model_inputs["input_ids"].shape[-1]
|
53 |
+
|
54 |
+
# Generate the response
|
55 |
+
with torch.inference_mode():
|
56 |
+
generation = model.generate(
|
57 |
+
**model_inputs, max_new_tokens=max_new_tokens, do_sample=False
|
58 |
+
)
|
59 |
+
generation = generation[0][input_len:]
|
60 |
+
decoded = processor.decode(generation, skip_special_tokens=True)
|
61 |
+
|
62 |
+
# Log the successful generation of the caption
|
63 |
+
logging.info("Caption generated successfully.")
|
64 |
+
|
65 |
+
# Return the generated caption
|
66 |
+
return decoded
|
67 |
+
|
68 |
+
# Handle exceptions that may occur during caption generation
|
69 |
+
except Exception as e:
|
70 |
+
# Custom exception handling
|
71 |
+
raise CustomExceptionHandling(e, sys) from e
|