Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -12,6 +12,7 @@ import torch
|
|
12 |
import numpy as np
|
13 |
from PIL import Image, ImageOps
|
14 |
import cv2
|
|
|
15 |
|
16 |
from transformers import (
|
17 |
Qwen2VLForConditionalGeneration,
|
@@ -35,32 +36,40 @@ MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
|
35 |
|
36 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
37 |
|
38 |
-
#
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
# Preprocessing functions for SmolDocling-256M
|
66 |
def add_random_padding(image, min_percent=0.1, max_percent=0.10):
|
@@ -71,7 +80,7 @@ def add_random_padding(image, min_percent=0.1, max_percent=0.10):
|
|
71 |
pad_h_percent = random.uniform(min_percent, max_percent)
|
72 |
pad_w = int(width * pad_w_percent)
|
73 |
pad_h = int(height * pad_h_percent)
|
74 |
-
corner_pixel = image.getpixel((0, 0))
|
75 |
padded_image = ImageOps.expand(image, border=(pad_w, pad_h, pad_w, pad_h), fill=corner_pixel)
|
76 |
return padded_image
|
77 |
|
@@ -109,11 +118,12 @@ def downsample_video(video_path):
|
|
109 |
return frames
|
110 |
|
111 |
# Dolphin-specific functions
|
|
|
112 |
def model_chat(prompt, image, is_batch=False):
|
113 |
"""Use Dolphin model for inference, supporting both single and batch processing."""
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
|
118 |
if not is_batch:
|
119 |
images = [image]
|
@@ -122,33 +132,30 @@ def model_chat(prompt, image, is_batch=False):
|
|
122 |
images = image
|
123 |
prompts = prompt if isinstance(prompt, list) else [prompt] * len(images)
|
124 |
|
125 |
-
inputs =
|
126 |
pixel_values = inputs.pixel_values.half()
|
127 |
|
128 |
prompts = [f"<s>{p} <Answer/>" for p in prompts]
|
129 |
-
prompt_inputs =
|
130 |
-
prompts,
|
131 |
-
add_special_tokens=False, # Explicitly set to False
|
132 |
-
return_tensors="pt",
|
133 |
-
padding=True
|
134 |
).to(device)
|
135 |
|
136 |
-
outputs =
|
137 |
pixel_values=pixel_values,
|
138 |
decoder_input_ids=prompt_inputs.input_ids,
|
139 |
decoder_attention_mask=prompt_inputs.attention_mask,
|
140 |
min_length=1,
|
141 |
max_length=4096,
|
142 |
-
pad_token_id=
|
143 |
-
eos_token_id=
|
144 |
use_cache=True,
|
145 |
-
bad_words_ids=[[
|
146 |
return_dict_in_generate=True,
|
147 |
do_sample=False,
|
148 |
num_beams=1,
|
149 |
repetition_penalty=1.1
|
150 |
)
|
151 |
-
sequences =
|
152 |
|
153 |
results = []
|
154 |
for i, sequence in enumerate(sequences):
|
@@ -157,6 +164,7 @@ def model_chat(prompt, image, is_batch=False):
|
|
157 |
|
158 |
return results[0] if not is_batch else results
|
159 |
|
|
|
160 |
def process_element_batch(elements, prompt, max_batch_size=16):
|
161 |
"""Process a batch of elements with the same prompt."""
|
162 |
results = []
|
@@ -244,24 +252,41 @@ def generate_markdown(recognition_results):
|
|
244 |
markdown += f"{element['text']}\n\n"
|
245 |
return markdown.strip()
|
246 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
247 |
def process_image_with_dolphin(image):
|
248 |
"""Process a single image with Dolphin model."""
|
249 |
-
|
250 |
-
|
|
|
|
|
|
|
251 |
markdown_content = generate_markdown(elements)
|
252 |
return markdown_content
|
253 |
|
254 |
@spaces.GPU
|
255 |
def generate_image(model_name: str, text: str, image: Image.Image,
|
256 |
-
max_new_tokens: int = 1024,
|
257 |
-
|
258 |
-
top_p: float = 0.9,
|
259 |
-
top_k: int = 50,
|
260 |
-
repetition_penalty: float = 1.2):
|
261 |
"""Generate responses for image input using the selected model."""
|
262 |
if model_name == "ByteDance-s-Dolphin":
|
263 |
if image is None:
|
264 |
-
yield "Please upload an image."
|
265 |
return
|
266 |
markdown_content = process_image_with_dolphin(image)
|
267 |
yield markdown_content
|
@@ -280,7 +305,10 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
280 |
yield "Please upload an image."
|
281 |
return
|
282 |
|
283 |
-
images = [image]
|
|
|
|
|
|
|
284 |
|
285 |
if model_name == "SmolDocling-256M-preview":
|
286 |
if "OTSL" in text or "code" in text:
|
@@ -334,11 +362,8 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
334 |
|
335 |
@spaces.GPU
|
336 |
def generate_video(model_name: str, text: str, video_path: str,
|
337 |
-
max_new_tokens: int = 1024,
|
338 |
-
|
339 |
-
top_p: float = 0.9,
|
340 |
-
top_k: int = 50,
|
341 |
-
repetition_penalty: float = 1.2):
|
342 |
"""Generate responses for video input using the selected model."""
|
343 |
if model_name == "ByteDance-s-Dolphin":
|
344 |
if video_path is None:
|
@@ -346,10 +371,10 @@ def generate_video(model_name: str, text: str, video_path: str,
|
|
346 |
return
|
347 |
frames = downsample_video(video_path)
|
348 |
markdown_contents = []
|
349 |
-
for frame, _ in frames:
|
350 |
markdown_content = process_image_with_dolphin(frame)
|
351 |
-
markdown_contents.append(markdown_content)
|
352 |
-
combined_markdown = "\n\n".join(markdown_contents)
|
353 |
yield combined_markdown
|
354 |
else:
|
355 |
if model_name == "olmOCR-7B-0225-preview":
|
@@ -419,7 +444,7 @@ def generate_video(model_name: str, text: str, video_path: str,
|
|
419 |
else:
|
420 |
yield cleaned_output
|
421 |
|
422 |
-
# Define examples
|
423 |
image_examples = [
|
424 |
["Convert this page to docling", "images/1.png"],
|
425 |
["OCR the image", "images/2.jpg"],
|
@@ -441,28 +466,23 @@ css = """
|
|
441 |
}
|
442 |
"""
|
443 |
|
444 |
-
# Create
|
445 |
with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
|
446 |
gr.Markdown("# **[Docling-VLMs](https://huggingface.co/collections/prithivMLmods/multimodal-implementations-67c9982ea04b39f0608badb0)**")
|
|
|
447 |
with gr.Row():
|
448 |
with gr.Column():
|
449 |
with gr.Tabs():
|
450 |
with gr.TabItem("Image Inference"):
|
451 |
image_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
|
452 |
-
image_upload = gr.Image(type="pil", label="Image")
|
453 |
image_submit = gr.Button("Submit", elem_classes="submit-btn")
|
454 |
-
gr.Examples(
|
455 |
-
examples=image_examples,
|
456 |
-
inputs=[image_query, image_upload]
|
457 |
-
)
|
458 |
with gr.TabItem("Video Inference"):
|
459 |
video_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
|
460 |
video_upload = gr.Video(label="Video")
|
461 |
video_submit = gr.Button("Submit", elem_classes="submit-btn")
|
462 |
-
gr.Examples(
|
463 |
-
examples=video_examples,
|
464 |
-
inputs=[video_query, video_upload]
|
465 |
-
)
|
466 |
with gr.Accordion("Advanced options", open=False):
|
467 |
max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS)
|
468 |
temperature = gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.6)
|
|
|
12 |
import numpy as np
|
13 |
from PIL import Image, ImageOps
|
14 |
import cv2
|
15 |
+
import pymupdf
|
16 |
|
17 |
from transformers import (
|
18 |
Qwen2VLForConditionalGeneration,
|
|
|
36 |
|
37 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
38 |
|
39 |
+
# Global variables for Dolphin model
|
40 |
+
model_k = None
|
41 |
+
processor_k = None
|
42 |
+
tokenizer_k = None
|
43 |
+
|
44 |
+
# Load models
|
45 |
+
def initialize_models():
|
46 |
+
global model_k, processor_k, tokenizer_k
|
47 |
+
# Load olmOCR-7B-0225-preview
|
48 |
+
MODEL_ID_M = "allenai/olmOCR-7B-0225-preview"
|
49 |
+
processor_m = AutoProcessor.from_pretrained(MODEL_ID_M, trust_remote_code=True)
|
50 |
+
model_m = Qwen2VLForConditionalGeneration.from_pretrained(
|
51 |
+
MODEL_ID_M, trust_remote_code=True, torch_dtype=torch.float16
|
52 |
+
).to(device).eval()
|
53 |
+
|
54 |
+
# Load ByteDance's Dolphin
|
55 |
+
MODEL_ID_K = "ByteDance/Dolphin"
|
56 |
+
processor_k = AutoProcessor.from_pretrained(MODEL_ID_K, trust_remote_code=True)
|
57 |
+
if model_k is None:
|
58 |
+
model_k = VisionEncoderDecoderModel.from_pretrained(
|
59 |
+
MODEL_ID_K, trust_remote_code=True, torch_dtype=torch.float16
|
60 |
+
).to(device).eval()
|
61 |
+
tokenizer_k = processor_k.tokenizer
|
62 |
+
|
63 |
+
# Load SmolDocling-256M-preview
|
64 |
+
MODEL_ID_X = "ds4sd/SmolDocling-256M-preview"
|
65 |
+
processor_x = AutoProcessor.from_pretrained(MODEL_ID_X, trust_remote_code=True)
|
66 |
+
model_x = AutoModelForVision2Seq.from_pretrained(
|
67 |
+
MODEL_ID_X, trust_remote_code=True, torch_dtype=torch.float16
|
68 |
+
).to(device).eval()
|
69 |
+
|
70 |
+
return processor_m, model_m, processor_x, model_x
|
71 |
+
|
72 |
+
processor_m, model_m, processor_x, model_x = initialize_models()
|
73 |
|
74 |
# Preprocessing functions for SmolDocling-256M
|
75 |
def add_random_padding(image, min_percent=0.1, max_percent=0.10):
|
|
|
80 |
pad_h_percent = random.uniform(min_percent, max_percent)
|
81 |
pad_w = int(width * pad_w_percent)
|
82 |
pad_h = int(height * pad_h_percent)
|
83 |
+
corner_pixel = image.getpixel((0, 0))
|
84 |
padded_image = ImageOps.expand(image, border=(pad_w, pad_h, pad_w, pad_h), fill=corner_pixel)
|
85 |
return padded_image
|
86 |
|
|
|
118 |
return frames
|
119 |
|
120 |
# Dolphin-specific functions
|
121 |
+
@spaces.GPU
|
122 |
def model_chat(prompt, image, is_batch=False):
|
123 |
"""Use Dolphin model for inference, supporting both single and batch processing."""
|
124 |
+
global model_k, processor_k, tokenizer_k
|
125 |
+
if model_k is None:
|
126 |
+
initialize_models()
|
127 |
|
128 |
if not is_batch:
|
129 |
images = [image]
|
|
|
132 |
images = image
|
133 |
prompts = prompt if isinstance(prompt, list) else [prompt] * len(images)
|
134 |
|
135 |
+
inputs = processor_k(images, return_tensors="pt", padding=True).to(device)
|
136 |
pixel_values = inputs.pixel_values.half()
|
137 |
|
138 |
prompts = [f"<s>{p} <Answer/>" for p in prompts]
|
139 |
+
prompt_inputs = tokenizer_k(
|
140 |
+
prompts, add_special_tokens=False, return_tensors="pt", padding=True
|
|
|
|
|
|
|
141 |
).to(device)
|
142 |
|
143 |
+
outputs = model_k.generate(
|
144 |
pixel_values=pixel_values,
|
145 |
decoder_input_ids=prompt_inputs.input_ids,
|
146 |
decoder_attention_mask=prompt_inputs.attention_mask,
|
147 |
min_length=1,
|
148 |
max_length=4096,
|
149 |
+
pad_token_id=tokenizer_k.pad_token_id,
|
150 |
+
eos_token_id=tokenizer_k.eos_token_id,
|
151 |
use_cache=True,
|
152 |
+
bad_words_ids=[[tokenizer_k.unk_token_id]],
|
153 |
return_dict_in_generate=True,
|
154 |
do_sample=False,
|
155 |
num_beams=1,
|
156 |
repetition_penalty=1.1
|
157 |
)
|
158 |
+
sequences = tokenizer_k.batch_decode(outputs.sequences, skip_special_tokens=False)
|
159 |
|
160 |
results = []
|
161 |
for i, sequence in enumerate(sequences):
|
|
|
164 |
|
165 |
return results[0] if not is_batch else results
|
166 |
|
167 |
+
@spaces.GPU
|
168 |
def process_element_batch(elements, prompt, max_batch_size=16):
|
169 |
"""Process a batch of elements with the same prompt."""
|
170 |
results = []
|
|
|
252 |
markdown += f"{element['text']}\n\n"
|
253 |
return markdown.strip()
|
254 |
|
255 |
+
def convert_to_image(image):
|
256 |
+
"""Convert uploaded file to PIL Image, handling PDFs by extracting the first page."""
|
257 |
+
if isinstance(image, str): # File path from Gradio
|
258 |
+
if image.lower().endswith('.pdf'):
|
259 |
+
doc = pymupdf.open(image)
|
260 |
+
page = doc[0]
|
261 |
+
pix = page.get_pixmap()
|
262 |
+
img_data = pix.tobytes("png")
|
263 |
+
pil_image = Image.open(io.BytesIO(img_data)).convert("RGB")
|
264 |
+
doc.close()
|
265 |
+
return pil_image
|
266 |
+
else:
|
267 |
+
return Image.open(image).convert("RGB")
|
268 |
+
elif isinstance(image, Image.Image): # Already a PIL Image
|
269 |
+
return image.convert("RGB")
|
270 |
+
return None
|
271 |
+
|
272 |
def process_image_with_dolphin(image):
|
273 |
"""Process a single image with Dolphin model."""
|
274 |
+
pil_image = convert_to_image(image)
|
275 |
+
if pil_image is None:
|
276 |
+
return "Error: Unable to process the uploaded file."
|
277 |
+
layout_output = model_chat("Parse the reading order of this document.", pil_image)
|
278 |
+
elements = process_elements(layout_output, pil_image)
|
279 |
markdown_content = generate_markdown(elements)
|
280 |
return markdown_content
|
281 |
|
282 |
@spaces.GPU
|
283 |
def generate_image(model_name: str, text: str, image: Image.Image,
|
284 |
+
max_new_tokens: int = 1024, temperature: float = 0.6,
|
285 |
+
top_p: float = 0.9, top_k: int = 50, repetition_penalty: float = 1.2):
|
|
|
|
|
|
|
286 |
"""Generate responses for image input using the selected model."""
|
287 |
if model_name == "ByteDance-s-Dolphin":
|
288 |
if image is None:
|
289 |
+
yield "Please upload an image or PDF (first page will be processed)."
|
290 |
return
|
291 |
markdown_content = process_image_with_dolphin(image)
|
292 |
yield markdown_content
|
|
|
305 |
yield "Please upload an image."
|
306 |
return
|
307 |
|
308 |
+
images = [convert_to_image(image)]
|
309 |
+
if images[0] is None:
|
310 |
+
yield "Error: Unable to process the uploaded file."
|
311 |
+
return
|
312 |
|
313 |
if model_name == "SmolDocling-256M-preview":
|
314 |
if "OTSL" in text or "code" in text:
|
|
|
362 |
|
363 |
@spaces.GPU
|
364 |
def generate_video(model_name: str, text: str, video_path: str,
|
365 |
+
max_new_tokens: int = 1024, temperature: float = 0.6,
|
366 |
+
top_p: float = 0.9, top_k: int = 50, repetition_penalty: float = 1.2):
|
|
|
|
|
|
|
367 |
"""Generate responses for video input using the selected model."""
|
368 |
if model_name == "ByteDance-s-Dolphin":
|
369 |
if video_path is None:
|
|
|
371 |
return
|
372 |
frames = downsample_video(video_path)
|
373 |
markdown_contents = []
|
374 |
+
for idx, (frame, _) in enumerate(frames):
|
375 |
markdown_content = process_image_with_dolphin(frame)
|
376 |
+
markdown_contents.append(f"**Frame {idx + 1}:**\n{markdown_content}")
|
377 |
+
combined_markdown = "\n\n---\n\n".join(markdown_contents)
|
378 |
yield combined_markdown
|
379 |
else:
|
380 |
if model_name == "olmOCR-7B-0225-preview":
|
|
|
444 |
else:
|
445 |
yield cleaned_output
|
446 |
|
447 |
+
# Define examples
|
448 |
image_examples = [
|
449 |
["Convert this page to docling", "images/1.png"],
|
450 |
["OCR the image", "images/2.jpg"],
|
|
|
466 |
}
|
467 |
"""
|
468 |
|
469 |
+
# Create Gradio Interface
|
470 |
with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
|
471 |
gr.Markdown("# **[Docling-VLMs](https://huggingface.co/collections/prithivMLmods/multimodal-implementations-67c9982ea04b39f0608badb0)**")
|
472 |
+
gr.Markdown("**Note:** For Dolphin model, the text query is ignored, and PDFs are processed by parsing the first page.")
|
473 |
with gr.Row():
|
474 |
with gr.Column():
|
475 |
with gr.Tabs():
|
476 |
with gr.TabItem("Image Inference"):
|
477 |
image_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
|
478 |
+
image_upload = gr.Image(type="pil", label="Image or PDF")
|
479 |
image_submit = gr.Button("Submit", elem_classes="submit-btn")
|
480 |
+
gr.Examples(examples=image_examples, inputs=[image_query, image_upload])
|
|
|
|
|
|
|
481 |
with gr.TabItem("Video Inference"):
|
482 |
video_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
|
483 |
video_upload = gr.Video(label="Video")
|
484 |
video_submit = gr.Button("Submit", elem_classes="submit-btn")
|
485 |
+
gr.Examples(examples=video_examples, inputs=[video_query, video_upload])
|
|
|
|
|
|
|
486 |
with gr.Accordion("Advanced options", open=False):
|
487 |
max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS)
|
488 |
temperature = gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.6)
|