DocUA commited on
Commit
f7c3947
·
1 Parent(s): 03c4021

feat: remove temperature parameter from UI and hardcode inference temperature to 0.0

Browse files
Files changed (2) hide show
  1. LightOnOCR-1B-Demo +1 -1
  2. hf_space/app.py +9 -16
LightOnOCR-1B-Demo CHANGED
@@ -1 +1 @@
1
- Subproject commit 625560a9e67d828b7561e830fb258b221ce29ee7
 
1
+ Subproject commit 394149f9369a36c1654e32e50356409c68447727
hf_space/app.py CHANGED
@@ -55,14 +55,15 @@ def process_pdf(pdf_path, num_pages=1, scale=2.0):
55
  return images, total_pages
56
 
57
  @spaces.GPU(duration=120) # Increase duration for OCR
58
- def run_inference(image, temperature, max_tokens):
59
  """Run inference on GPU."""
60
  global BACKEND
61
  if BACKEND is None:
62
  load_backend()
63
- return BACKEND.process_image(image, temperature=temperature, max_tokens=max_tokens)
64
 
65
- def process_input(file_input, scale, temperature, max_tokens, num_pages):
 
66
  """Process uploaded file with OCR."""
67
  if file_input is None:
68
  yield "Idle", "Please upload an image or PDF first.", "", "", None
@@ -71,7 +72,8 @@ def process_input(file_input, scale, temperature, max_tokens, num_pages):
71
  images_to_process = []
72
  page_info = ""
73
  display_image = None
74
-
 
75
  file_path = Path(file_input) if isinstance(file_input, str) else Path(file_input.name)
76
  if not file_path.exists():
77
  yield "Error", f"File not accessible: {file_path}", "", "", None
@@ -107,8 +109,8 @@ def process_input(file_input, scale, temperature, max_tokens, num_pages):
107
  for i, img in enumerate(images_to_process):
108
  try:
109
  print(f"Processing page {i+1}/{len(images_to_process)}...")
110
- # Run inference on GPU
111
- text = run_inference(img, temperature=temperature, max_tokens=max_tokens)
112
  all_texts.append(text.strip())
113
 
114
  # Update progress
@@ -179,15 +181,6 @@ with gr.Blocks(title="📖 LightOnOCR-1B Demo", theme=gr.themes.Soft()) as demo:
179
  label="PDF Pages",
180
  info="Number of pages to process (max 10)"
181
  )
182
-
183
- temperature = gr.Slider(
184
- minimum=0.0,
185
- maximum=1.0,
186
- value=0.1,
187
- step=0.05,
188
- label="Temperature",
189
- info="0 = deterministic"
190
- )
191
 
192
  page_info = gr.Textbox(
193
  label="Processing Info",
@@ -221,7 +214,7 @@ with gr.Blocks(title="📖 LightOnOCR-1B Demo", theme=gr.themes.Soft()) as demo:
221
  # Event handlers
222
  submit_btn.click(
223
  fn=process_input,
224
- inputs=[file_input, scale_slider, temperature, max_tokens_slider, num_pages],
225
  outputs=[status_display, output_text, raw_output, page_info, rendered_image]
226
  )
227
 
 
55
  return images, total_pages
56
 
57
  @spaces.GPU(duration=120) # Increase duration for OCR
58
+ def run_inference(image, max_tokens):
59
  """Run inference on GPU."""
60
  global BACKEND
61
  if BACKEND is None:
62
  load_backend()
63
+ return BACKEND.process_image(image, temperature=0.0, max_tokens=max_tokens)
64
 
65
+
66
+ def process_input(file_input, scale, max_tokens, num_pages):
67
  """Process uploaded file with OCR."""
68
  if file_input is None:
69
  yield "Idle", "Please upload an image or PDF first.", "", "", None
 
72
  images_to_process = []
73
  page_info = ""
74
  display_image = None
75
+
76
+ # ... (rest of image loading logic same as before, simplified for diff clarity)
77
  file_path = Path(file_input) if isinstance(file_input, str) else Path(file_input.name)
78
  if not file_path.exists():
79
  yield "Error", f"File not accessible: {file_path}", "", "", None
 
109
  for i, img in enumerate(images_to_process):
110
  try:
111
  print(f"Processing page {i+1}/{len(images_to_process)}...")
112
+ # Run inference on GPU (hardcoded temp=0.0)
113
+ text = run_inference(img, max_tokens=max_tokens)
114
  all_texts.append(text.strip())
115
 
116
  # Update progress
 
181
  label="PDF Pages",
182
  info="Number of pages to process (max 10)"
183
  )
 
 
 
 
 
 
 
 
 
184
 
185
  page_info = gr.Textbox(
186
  label="Processing Info",
 
214
  # Event handlers
215
  submit_btn.click(
216
  fn=process_input,
217
+ inputs=[file_input, scale_slider, max_tokens_slider, num_pages],
218
  outputs=[status_display, output_text, raw_output, page_info, rendered_image]
219
  )
220