Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,658 +1,149 @@
|
|
|
|
1 |
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
import os
|
6 |
-
|
7 |
-
import time
|
8 |
-
|
9 |
from threading import Thread
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
import gradio as gr
|
14 |
-
|
15 |
import spaces
|
16 |
-
|
17 |
from PIL import Image
|
18 |
-
|
19 |
import torch
|
|
|
|
|
|
|
|
|
20 |
|
21 |
-
|
22 |
-
|
23 |
-
AutoProcessor,
|
24 |
-
|
25 |
-
AutoModelForImageTextToText,
|
26 |
-
|
27 |
-
Qwen2_5_VLForConditionalGeneration,
|
28 |
-
|
29 |
-
)
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
# ---------------------------
|
34 |
-
|
35 |
-
# Models
|
36 |
-
|
37 |
-
# ---------------------------
|
38 |
-
|
39 |
MODEL_PATHS = {
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
"prithivMLmods/Qwen2.5-VL-7B-Abliterated-Caption-it",
|
44 |
-
|
45 |
-
Qwen2_5_VLForConditionalGeneration,
|
46 |
-
|
47 |
-
),
|
48 |
-
|
49 |
-
"Model 2 (simple and scanned handwritting )": (
|
50 |
-
|
51 |
-
"nanonets/Nanonets-OCR-s",
|
52 |
-
|
53 |
-
Qwen2_5_VLForConditionalGeneration,
|
54 |
-
|
55 |
-
),
|
56 |
-
|
57 |
-
"Model 3 (structured handwritting)": (
|
58 |
-
|
59 |
-
"Emeritus-21/Finetuned-full-HTR-model",
|
60 |
-
|
61 |
-
AutoModelForImageTextToText,
|
62 |
-
|
63 |
-
),
|
64 |
-
|
65 |
}
|
66 |
|
67 |
-
|
68 |
-
|
69 |
MAX_NEW_TOKENS_DEFAULT = 512
|
70 |
-
|
71 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
72 |
|
73 |
-
|
74 |
-
|
75 |
-
# ---------------------------
|
76 |
-
|
77 |
-
# Preload models at startup
|
78 |
-
|
79 |
-
# ---------------------------
|
80 |
-
|
81 |
-
_loaded_processors = {}
|
82 |
-
|
83 |
-
_loaded_models = {}
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
print("🚀 Preloading models into GPU/CPU memory...")
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
for name, (repo_id, cls) in MODEL_PATHS.items():
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
trust_remote_code=True,
|
104 |
-
|
105 |
-
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
106 |
-
|
107 |
-
low_cpu_mem_usage=True,
|
108 |
-
|
109 |
-
).to(device).eval()
|
110 |
-
|
111 |
-
_loaded_processors[name] = processor
|
112 |
-
|
113 |
-
_loaded_models[name] = model
|
114 |
-
|
115 |
-
print(f"✅ {name} ready.")
|
116 |
-
|
117 |
-
except Exception as e:
|
118 |
-
|
119 |
-
print(f"⚠️ Failed to load {name}: {e}")
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
# ---------------------------
|
124 |
-
|
125 |
-
# Warmup (GPU)
|
126 |
-
|
127 |
-
# ---------------------------
|
128 |
-
|
129 |
@spaces.GPU
|
130 |
-
|
131 |
def warmup(progress=gr.Progress(track_tqdm=True)):
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
if tokenizer and hasattr(tokenizer, "apply_chat_template"):
|
148 |
-
|
149 |
-
chat_prompt = tokenizer.apply_chat_template(
|
150 |
-
|
151 |
-
messages, tokenize=False, add_generation_prompt=True
|
152 |
-
|
153 |
-
)
|
154 |
-
|
155 |
-
else:
|
156 |
-
|
157 |
-
chat_prompt = "Warmup."
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
inputs = processor(
|
162 |
-
|
163 |
-
text=[chat_prompt],
|
164 |
-
|
165 |
-
images=None,
|
166 |
-
|
167 |
-
return_tensors="pt"
|
168 |
-
|
169 |
-
).to(device)
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
with torch.inference_mode():
|
174 |
-
|
175 |
-
_ = model.generate(**inputs, max_new_tokens=1)
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
return f"GPU warm and {default_model_choice} ready."
|
180 |
-
|
181 |
-
except Exception as e:
|
182 |
-
|
183 |
-
return f"Warmup skipped: {e}"
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
# ---------------------------
|
188 |
-
|
189 |
-
# Helpers
|
190 |
-
|
191 |
-
# ---------------------------
|
192 |
-
|
193 |
def _build_inputs(processor, tokenizer, image: Image.Image, prompt: str):
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
{
|
200 |
-
|
201 |
-
"role": "user",
|
202 |
-
|
203 |
-
"content": [
|
204 |
-
|
205 |
-
{"type": "image", "image": image},
|
206 |
-
|
207 |
-
{"type": "text", "text": prompt},
|
208 |
-
|
209 |
-
],
|
210 |
-
|
211 |
-
}
|
212 |
-
|
213 |
-
]
|
214 |
-
|
215 |
-
if tokenizer and hasattr(tokenizer, "apply_chat_template"):
|
216 |
-
|
217 |
-
chat_prompt = tokenizer.apply_chat_template(
|
218 |
-
|
219 |
-
messages, tokenize=False, add_generation_prompt=True
|
220 |
-
|
221 |
-
)
|
222 |
-
|
223 |
-
return processor(text=[chat_prompt], images=[image], return_tensors="pt")
|
224 |
-
|
225 |
-
# Fallback: plain prompt + image
|
226 |
-
|
227 |
-
return processor(text=[prompt], images=[image], return_tensors="pt")
|
228 |
-
|
229 |
-
|
230 |
|
231 |
def _decode_text(model, processor, tokenizer, output_ids):
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
try:
|
238 |
-
|
239 |
-
if hasattr(processor, "batch_decode"):
|
240 |
-
|
241 |
-
text = processor.batch_decode(output_ids, skip_special_tokens=True)[0]
|
242 |
-
|
243 |
-
return text
|
244 |
-
|
245 |
-
except Exception:
|
246 |
-
|
247 |
-
pass
|
248 |
-
|
249 |
-
try:
|
250 |
-
|
251 |
-
if tokenizer is not None:
|
252 |
-
|
253 |
-
text = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0]
|
254 |
-
|
255 |
-
return text
|
256 |
-
|
257 |
-
except Exception:
|
258 |
-
|
259 |
-
pass
|
260 |
-
|
261 |
-
try:
|
262 |
-
|
263 |
-
model_tok = getattr(model, "tokenizer", None)
|
264 |
-
|
265 |
-
if model_tok is not None:
|
266 |
-
|
267 |
-
text = model_tok.batch_decode(output_ids, skip_special_tokens=True)[0]
|
268 |
-
|
269 |
-
return text
|
270 |
-
|
271 |
-
except Exception:
|
272 |
-
|
273 |
-
pass
|
274 |
-
|
275 |
-
# Last-resort string
|
276 |
-
|
277 |
-
return str(output_ids)
|
278 |
-
|
279 |
-
|
280 |
|
281 |
def _default_prompt(query: str | None) -> str:
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
|
288 |
-
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
"- Preserve original structure and line breaks.\n"
|
294 |
-
|
295 |
-
"- Keep spacing, bullet points, numbering, and indentation.\n"
|
296 |
-
|
297 |
-
"- Render tables as Markdown tables if present.\n"
|
298 |
-
|
299 |
-
"- Do NOT autocorrect spelling or grammar.\n"
|
300 |
-
|
301 |
-
"- Do NOT merge lines.\n"
|
302 |
-
|
303 |
-
"Return RAW transcription only."
|
304 |
-
|
305 |
-
)
|
306 |
-
|
307 |
-
|
308 |
-
|
309 |
-
# ---------------------------
|
310 |
-
|
311 |
-
# OCR Function (NO STREAMING / NO yield) ✅ FIX
|
312 |
-
|
313 |
-
# ---------------------------
|
314 |
-
|
315 |
@spaces.GPU
|
316 |
-
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
|
329 |
-
|
330 |
-
|
331 |
-
|
332 |
-
|
333 |
-
repetition_penalty: float = 1.0,
|
334 |
-
|
335 |
-
progress=gr.Progress(track_tqdm=True),
|
336 |
-
|
337 |
-
):
|
338 |
-
|
339 |
-
if image is None:
|
340 |
-
|
341 |
-
return "Please upload or capture an image."
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
-
if model_choice not in _loaded_models:
|
346 |
-
|
347 |
-
return f"Invalid model: {model_choice}"
|
348 |
-
|
349 |
-
|
350 |
-
|
351 |
-
processor = _loaded_processors[model_choice]
|
352 |
-
|
353 |
-
model = _loaded_models[model_choice]
|
354 |
-
|
355 |
-
tokenizer = getattr(processor, "tokenizer", None)
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
prompt = _default_prompt(query)
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
# Build inputs
|
364 |
-
|
365 |
-
batch = _build_inputs(processor, tokenizer, image, prompt).to(device)
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
# Generate (no streaming)
|
370 |
-
|
371 |
-
with torch.inference_mode():
|
372 |
-
|
373 |
-
output_ids = model.generate(
|
374 |
-
|
375 |
-
**batch,
|
376 |
-
|
377 |
-
max_new_tokens=max_new_tokens,
|
378 |
-
|
379 |
-
do_sample=False,
|
380 |
-
|
381 |
-
temperature=temperature,
|
382 |
-
|
383 |
-
top_p=top_p,
|
384 |
-
|
385 |
-
top_k=top_k,
|
386 |
-
|
387 |
-
repetition_penalty=repetition_penalty,
|
388 |
-
|
389 |
-
)
|
390 |
-
|
391 |
-
|
392 |
-
|
393 |
-
# Decode
|
394 |
-
|
395 |
-
decoded = _decode_text(model, processor, tokenizer, output_ids)
|
396 |
-
|
397 |
-
cleaned = decoded.replace("<|im_end|>", "").strip()
|
398 |
-
|
399 |
-
return cleaned
|
400 |
-
|
401 |
-
|
402 |
-
|
403 |
-
# ---------------------------
|
404 |
-
|
405 |
-
# Export Helpers
|
406 |
-
|
407 |
-
# ---------------------------
|
408 |
-
|
409 |
-
from reportlab.platypus import SimpleDocTemplate, Paragraph
|
410 |
-
|
411 |
-
from reportlab.lib.styles import getSampleStyleSheet
|
412 |
-
|
413 |
-
from docx import Document
|
414 |
-
|
415 |
-
|
416 |
-
|
417 |
-
def _safe_text(text: str) -> str:
|
418 |
-
|
419 |
-
return (text or "").strip()
|
420 |
-
|
421 |
-
|
422 |
|
423 |
def save_as_pdf(text):
|
424 |
-
|
425 |
-
|
426 |
-
|
427 |
-
|
428 |
-
|
429 |
-
|
430 |
-
|
431 |
-
filepath = "output.pdf"
|
432 |
-
|
433 |
-
doc = SimpleDocTemplate(filepath)
|
434 |
-
|
435 |
-
styles = getSampleStyleSheet()
|
436 |
-
|
437 |
-
flowables = [Paragraph(t, styles["Normal"]) for t in text.splitlines() if t != ""]
|
438 |
-
|
439 |
-
if not flowables:
|
440 |
-
|
441 |
-
flowables = [Paragraph(" ", styles["Normal"])]
|
442 |
-
|
443 |
-
doc.build(flowables)
|
444 |
-
|
445 |
-
return filepath
|
446 |
-
|
447 |
-
|
448 |
|
449 |
def save_as_word(text):
|
450 |
-
|
451 |
-
|
452 |
-
|
453 |
-
|
454 |
-
|
455 |
-
|
456 |
-
|
457 |
-
filepath = "output.docx"
|
458 |
-
|
459 |
-
doc = Document()
|
460 |
-
|
461 |
-
for line in text.splitlines():
|
462 |
-
|
463 |
-
doc.add_paragraph(line)
|
464 |
-
|
465 |
-
doc.save(filepath)
|
466 |
-
|
467 |
-
return filepath
|
468 |
-
|
469 |
-
|
470 |
-
|
471 |
-
# gTTS uses Google TTS (requires outbound internet). Wrap in try/except so Space doesn't crash.
|
472 |
|
473 |
def save_as_audio(text):
|
|
|
|
|
|
|
|
|
474 |
|
475 |
-
|
476 |
-
|
477 |
-
if not text:
|
478 |
-
|
479 |
-
return None
|
480 |
-
|
481 |
-
try:
|
482 |
-
|
483 |
-
from gTTS import gTTS
|
484 |
-
|
485 |
-
filepath = "output.mp3"
|
486 |
-
|
487 |
-
tts = gTTS(text)
|
488 |
-
|
489 |
-
tts.save(filepath)
|
490 |
-
|
491 |
-
return filepath
|
492 |
-
|
493 |
-
except Exception as e:
|
494 |
-
|
495 |
-
print(f"gTTS failed: {e}")
|
496 |
-
|
497 |
-
return None
|
498 |
-
|
499 |
-
|
500 |
-
|
501 |
-
# ---------------------------
|
502 |
-
|
503 |
-
# Gradio Interface
|
504 |
-
|
505 |
-
# ---------------------------
|
506 |
-
|
507 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
508 |
-
|
509 |
-
|
510 |
-
|
511 |
-
|
512 |
-
|
513 |
-
|
514 |
-
|
515 |
-
|
516 |
-
|
517 |
-
|
518 |
-
|
519 |
-
|
520 |
-
|
521 |
-
|
522 |
-
|
523 |
-
|
524 |
-
|
525 |
-
|
526 |
-
|
527 |
-
|
528 |
-
|
529 |
-
|
530 |
-
|
531 |
-
|
532 |
-
|
533 |
-
)
|
534 |
-
|
535 |
-
|
536 |
-
|
537 |
-
# Upload + Webcam (Gradio 4.x uses `sources`)
|
538 |
-
|
539 |
-
image_input = gr.Image(
|
540 |
-
|
541 |
-
type="pil",
|
542 |
-
|
543 |
-
label="Upload / Capture Handwritten Image",
|
544 |
-
|
545 |
-
sources=["upload", "webcam"],
|
546 |
-
|
547 |
-
)
|
548 |
-
|
549 |
-
|
550 |
-
|
551 |
-
with gr.Accordion("⚙️ Advanced Options", open=False):
|
552 |
-
|
553 |
-
max_new_tokens = gr.Slider(1, 2048, value=MAX_NEW_TOKENS_DEFAULT, step=1, label="Max new tokens")
|
554 |
-
|
555 |
-
temperature = gr.Slider(0.1, 2.0, value=0.1, step=0.05, label="Temperature")
|
556 |
-
|
557 |
-
top_p = gr.Slider(0.05, 1.0, value=1.0, step=0.05, label="Top-p (nucleus)")
|
558 |
-
|
559 |
-
top_k = gr.Slider(0, 1000, value=0, step=1, label="Top-k")
|
560 |
-
|
561 |
-
repetition_penalty = gr.Slider(0.8, 2.0, value=1.0, step=0.05, label="Repetition penalty")
|
562 |
-
|
563 |
-
|
564 |
-
|
565 |
-
with gr.Row():
|
566 |
-
|
567 |
-
extract_btn = gr.Button("📤 Extract RAW Text", variant="primary")
|
568 |
-
|
569 |
-
clear_btn = gr.Button("🧹 Clear")
|
570 |
-
|
571 |
-
|
572 |
-
|
573 |
-
raw_output = gr.Textbox(
|
574 |
-
|
575 |
-
label="📜 RAW Structured Output (exact as written)",
|
576 |
-
|
577 |
-
lines=18,
|
578 |
-
|
579 |
-
show_copy_button=True,
|
580 |
-
|
581 |
-
)
|
582 |
-
|
583 |
-
|
584 |
-
|
585 |
-
with gr.Row():
|
586 |
-
|
587 |
-
pdf_btn = gr.Button("⬇️ Download as PDF")
|
588 |
-
|
589 |
-
word_btn = gr.Button("⬇️ Download as Word")
|
590 |
-
|
591 |
-
audio_btn = gr.Button("🔊 Download as Audio")
|
592 |
-
|
593 |
-
|
594 |
-
|
595 |
-
pdf_file = gr.File(label="PDF File")
|
596 |
-
|
597 |
-
word_file = gr.File(label="Word File")
|
598 |
-
|
599 |
-
audio_file = gr.File(label="Audio File")
|
600 |
-
|
601 |
-
|
602 |
-
|
603 |
-
extract_btn.click(
|
604 |
-
|
605 |
-
fn=ocr_image,
|
606 |
-
|
607 |
-
inputs=[
|
608 |
-
|
609 |
-
image_input,
|
610 |
-
|
611 |
-
model_choice,
|
612 |
-
|
613 |
-
query_input,
|
614 |
-
|
615 |
-
max_new_tokens,
|
616 |
-
|
617 |
-
temperature,
|
618 |
-
|
619 |
-
top_p,
|
620 |
-
|
621 |
-
top_k,
|
622 |
-
|
623 |
-
repetition_penalty,
|
624 |
-
|
625 |
-
],
|
626 |
-
|
627 |
-
outputs=[raw_output],
|
628 |
-
|
629 |
-
api_name="ocr_image",
|
630 |
-
|
631 |
-
)
|
632 |
-
|
633 |
-
|
634 |
-
|
635 |
-
pdf_btn.click(fn=save_as_pdf, inputs=[raw_output], outputs=[pdf_file])
|
636 |
-
|
637 |
-
word_btn.click(fn=save_as_word, inputs=[raw_output], outputs=[word_file])
|
638 |
-
|
639 |
-
audio_btn.click(fn=save_as_audio, inputs=[raw_output], outputs=[audio_file])
|
640 |
-
|
641 |
-
|
642 |
-
|
643 |
-
clear_btn.click(
|
644 |
-
|
645 |
-
fn=lambda: ("", None, "", MAX_NEW_TOKENS_DEFAULT, 0.1, 1.0, 0, 1.0),
|
646 |
-
|
647 |
-
outputs=[raw_output, image_input, query_input, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
|
648 |
-
|
649 |
-
)
|
650 |
-
|
651 |
-
|
652 |
|
653 |
if __name__ == "__main__":
|
654 |
-
|
655 |
-
# Keep queue for GPU tasks; limit concurrency for stability.
|
656 |
-
|
657 |
-
demo.queue(max_size=50).launch(show_error=True)
|
658 |
-
|
|
|
1 |
+
# app.py — HTR Space (Compact Version)
|
2 |
|
3 |
+
import os, time
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
from threading import Thread
|
|
|
|
|
|
|
5 |
import gradio as gr
|
|
|
6 |
import spaces
|
|
|
7 |
from PIL import Image
|
|
|
8 |
import torch
|
9 |
+
from transformers import AutoProcessor, AutoModelForImageTextToText, Qwen2_5_VLForConditionalGeneration
|
10 |
+
from reportlab.platypus import SimpleDocTemplate, Paragraph
|
11 |
+
from reportlab.lib.styles import getSampleStyleSheet
|
12 |
+
from docx import Document
|
13 |
|
14 |
+
# ---------------- Models ----------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
MODEL_PATHS = {
|
16 |
+
"Model 1 (Complex handwrittings )": ("prithivMLmods/Qwen2.5-VL-7B-Abliterated-Caption-it", Qwen2_5_VLForConditionalGeneration),
|
17 |
+
"Model 2 (simple and scanned handwritting )": ("nanonets/Nanonets-OCR-s", Qwen2_5_VLForConditionalGeneration),
|
18 |
+
"Model 3 (structured handwritting)": ("Emeritus-21/Finetuned-full-HTR-model", AutoModelForImageTextToText),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
}
|
20 |
|
|
|
|
|
21 |
MAX_NEW_TOKENS_DEFAULT = 512
|
|
|
22 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
23 |
|
24 |
+
_loaded_processors, _loaded_models = {}, {}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
print("🚀 Preloading models into GPU/CPU memory...")
|
|
|
|
|
|
|
26 |
for name, (repo_id, cls) in MODEL_PATHS.items():
|
27 |
+
try:
|
28 |
+
processor = AutoProcessor.from_pretrained(repo_id, trust_remote_code=True)
|
29 |
+
model = cls.from_pretrained(repo_id, trust_remote_code=True,
|
30 |
+
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
31 |
+
low_cpu_mem_usage=True).to(device).eval()
|
32 |
+
_loaded_processors[name], _loaded_models[name] = processor, model
|
33 |
+
print(f"✅ {name} ready.")
|
34 |
+
except Exception as e:
|
35 |
+
print(f"⚠️ Failed to load {name}: {e}")
|
36 |
+
|
37 |
+
# ---------------- GPU Warmup ----------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
@spaces.GPU
|
|
|
39 |
def warmup(progress=gr.Progress(track_tqdm=True)):
|
40 |
+
try:
|
41 |
+
default_model_choice = next(iter(MODEL_PATHS.keys()))
|
42 |
+
processor = _loaded_processors[default_model_choice]
|
43 |
+
model = _loaded_models[default_model_choice]
|
44 |
+
tokenizer = getattr(processor, "tokenizer", None)
|
45 |
+
messages = [{"role": "user", "content": [{"type": "text", "text": "Warmup."}]}]
|
46 |
+
chat_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) if tokenizer and hasattr(tokenizer, "apply_chat_template") else "Warmup."
|
47 |
+
inputs = processor(text=[chat_prompt], images=None, return_tensors="pt").to(device)
|
48 |
+
with torch.inference_mode(): _ = model.generate(**inputs, max_new_tokens=1)
|
49 |
+
return f"GPU warm and {default_model_choice} ready."
|
50 |
+
except Exception as e:
|
51 |
+
return f"Warmup skipped: {e}"
|
52 |
+
|
53 |
+
# ---------------- Helpers ----------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
def _build_inputs(processor, tokenizer, image: Image.Image, prompt: str):
|
55 |
+
messages = [{"role": "user", "content": [{"type": "image", "image": image}, {"type": "text", "text": prompt}]}]
|
56 |
+
if tokenizer and hasattr(tokenizer, "apply_chat_template"):
|
57 |
+
chat_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
58 |
+
return processor(text=[chat_prompt], images=[image], return_tensors="pt")
|
59 |
+
return processor(text=[prompt], images=[image], return_tensors="pt")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
def _decode_text(model, processor, tokenizer, output_ids):
|
62 |
+
for obj in [processor, tokenizer, getattr(model, "tokenizer", None)]:
|
63 |
+
try: return obj.batch_decode(output_ids, skip_special_tokens=True)[0]
|
64 |
+
except Exception: pass
|
65 |
+
return str(output_ids)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
67 |
def _default_prompt(query: str | None) -> str:
|
68 |
+
if query and query.strip(): return query.strip()
|
69 |
+
return ("You are a professional Handwritten OCR system.\n"
|
70 |
+
"TASK: Read the handwritten image and transcribe the text EXACTLY as written.\n"
|
71 |
+
"- Preserve original structure and line breaks.\n"
|
72 |
+
"- Keep spacing, bullet points, numbering, and indentation.\n"
|
73 |
+
"- Render tables as Markdown tables if present.\n"
|
74 |
+
"- Do NOT autocorrect spelling or grammar.\n"
|
75 |
+
"- Do NOT merge lines.\n"
|
76 |
+
"Return RAW transcription only.")
|
77 |
+
|
78 |
+
# ---------------- OCR Function ----------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
@spaces.GPU
|
80 |
+
def ocr_image(image: Image.Image, model_choice: str, query: str = None,
|
81 |
+
max_new_tokens: int = MAX_NEW_TOKENS_DEFAULT,
|
82 |
+
temperature: float = 0.1, top_p: float = 1.0, top_k: int = 0, repetition_penalty: float = 1.0,
|
83 |
+
progress=gr.Progress(track_tqdm=True)):
|
84 |
+
if image is None: return "Please upload or capture an image."
|
85 |
+
if model_choice not in _loaded_models: return f"Invalid model: {model_choice}"
|
86 |
+
processor, model, tokenizer = _loaded_processors[model_choice], _loaded_models[model_choice], getattr(_loaded_processors[model_choice], "tokenizer", None)
|
87 |
+
prompt = _default_prompt(query)
|
88 |
+
batch = _build_inputs(processor, tokenizer, image, prompt).to(device)
|
89 |
+
with torch.inference_mode():
|
90 |
+
output_ids = model.generate(**batch, max_new_tokens=max_new_tokens, do_sample=False,
|
91 |
+
temperature=temperature, top_p=top_p, top_k=top_k, repetition_penalty=repetition_penalty)
|
92 |
+
return _decode_text(model, processor, tokenizer, output_ids).replace("<|im_end|>", "").strip()
|
93 |
+
|
94 |
+
# ---------------- Export Helpers ----------------
|
95 |
+
def _safe_text(text: str) -> str: return (text or "").strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
|
97 |
def save_as_pdf(text):
|
98 |
+
text = _safe_text(text)
|
99 |
+
if not text: return None
|
100 |
+
doc = SimpleDocTemplate("output.pdf")
|
101 |
+
flowables = [Paragraph(t, getSampleStyleSheet()["Normal"]) for t in text.splitlines() if t != ""]
|
102 |
+
if not flowables: flowables = [Paragraph(" ", getSampleStyleSheet()["Normal"])]
|
103 |
+
doc.build(flowables)
|
104 |
+
return "output.pdf"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
|
106 |
def save_as_word(text):
|
107 |
+
text = _safe_text(text)
|
108 |
+
if not text: return None
|
109 |
+
doc = Document()
|
110 |
+
for line in text.splitlines(): doc.add_paragraph(line)
|
111 |
+
doc.save("output.docx")
|
112 |
+
return "output.docx"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
|
114 |
def save_as_audio(text):
|
115 |
+
text = _safe_text(text)
|
116 |
+
if not text: return None
|
117 |
+
try: from gTTS import gTTS; tts = gTTS(text); tts.save("output.mp3"); return "output.mp3"
|
118 |
+
except Exception as e: print(f"gTTS failed: {e}"); return None
|
119 |
|
120 |
+
# ---------------- Gradio Interface ----------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
121 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
122 |
+
gr.Markdown("## ✍🏾 wilson Handwritten OCR")
|
123 |
+
model_choice = gr.Radio(choices=list(MODEL_PATHS.keys()), value=list(MODEL_PATHS.keys())[0], label="Select OCR Model")
|
124 |
+
with gr.Tab("🖼 Image Inference"):
|
125 |
+
query_input = gr.Textbox(label="Custom Prompt (optional)", placeholder="Leave empty for RAW structured output")
|
126 |
+
image_input = gr.Image(type="pil", label="Upload / Capture Handwritten Image", sources=["upload", "webcam"])
|
127 |
+
with gr.Accordion("⚙️ Advanced Options", open=False):
|
128 |
+
max_new_tokens = gr.Slider(1, 2048, value=MAX_NEW_TOKENS_DEFAULT, step=1, label="Max new tokens")
|
129 |
+
temperature = gr.Slider(0.1, 2.0, value=0.1, step=0.05, label="Temperature")
|
130 |
+
top_p = gr.Slider(0.05, 1.0, value=1.0, step=0.05, label="Top-p (nucleus)")
|
131 |
+
top_k = gr.Slider(0, 1000, value=0, step=1, label="Top-k")
|
132 |
+
repetition_penalty = gr.Slider(0.8, 2.0, value=1.0, step=0.05, label="Repetition penalty")
|
133 |
+
extract_btn = gr.Button("📤 Extract RAW Text", variant="primary")
|
134 |
+
clear_btn = gr.Button("🧹 Clear")
|
135 |
+
raw_output = gr.Textbox(label="📜 RAW Structured Output (exact as written)", lines=18, show_copy_button=True)
|
136 |
+
pdf_btn = gr.Button("⬇️ Download as PDF")
|
137 |
+
word_btn = gr.Button("⬇️ Download as Word")
|
138 |
+
audio_btn = gr.Button("🔊 Download as Audio")
|
139 |
+
pdf_file, word_file, audio_file = gr.File(label="PDF File"), gr.File(label="Word File"), gr.File(label="Audio File")
|
140 |
+
|
141 |
+
extract_btn.click(fn=ocr_image, inputs=[image_input, model_choice, query_input, max_new_tokens, temperature, top_p, top_k, repetition_penalty], outputs=[raw_output], api_name="ocr_image")
|
142 |
+
pdf_btn.click(fn=save_as_pdf, inputs=[raw_output], outputs=[pdf_file])
|
143 |
+
word_btn.click(fn=save_as_word, inputs=[raw_output], outputs=[word_file])
|
144 |
+
audio_btn.click(fn=save_as_audio, inputs=[raw_output], outputs=[audio_file])
|
145 |
+
clear_btn.click(fn=lambda: ("", None, "", MAX_NEW_TOKENS_DEFAULT, 0.1, 1.0, 0, 1.0),
|
146 |
+
outputs=[raw_output, image_input, query_input, max_new_tokens, temperature, top_p, top_k, repetition_penalty])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
147 |
|
148 |
if __name__ == "__main__":
|
149 |
+
demo.queue(max_size=50).launch(show_error=True)
|
|
|
|
|
|
|
|