davanstrien HF Staff commited on
Commit
5951139
·
1 Parent(s): 1825db8

remove flashinfer

Browse files
Files changed (1) hide show
  1. nanonets-ocr.py +49 -59
nanonets-ocr.py CHANGED
@@ -5,14 +5,11 @@
5
  # "huggingface-hub[hf_transfer]",
6
  # "pillow",
7
  # "vllm",
8
- # "flashinfer-python",
9
  # "tqdm",
10
  # "toolz",
11
  # "torch", # Added for CUDA check
12
  # ]
13
- #
14
- # [[tool.uv.index]]
15
- # url = "https://flashinfer.ai/whl/cu121/torch2.4/"
16
  # ///
17
 
18
  """
@@ -72,12 +69,12 @@ def make_ocr_message(
72
  pil_img = Image.open(image)
73
  else:
74
  raise ValueError(f"Unsupported image type: {type(image)}")
75
-
76
  # Convert to base64 data URI
77
  buf = io.BytesIO()
78
  pil_img.save(buf, format="PNG")
79
  data_uri = f"data:image/png;base64,{base64.b64encode(buf.getvalue()).decode()}"
80
-
81
  # Return message in vLLM format
82
  return [
83
  {
@@ -105,31 +102,33 @@ def main(
105
  private: bool = False,
106
  ):
107
  """Process images from HF dataset through OCR model."""
108
-
109
  # Check CUDA availability first
110
  check_cuda_availability()
111
-
112
  # Enable HF_TRANSFER for faster downloads
113
  os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
114
-
115
  # Login to HF if token provided
116
  HF_TOKEN = hf_token or os.environ.get("HF_TOKEN")
117
  if HF_TOKEN:
118
  login(token=HF_TOKEN)
119
-
120
  # Load dataset
121
  logger.info(f"Loading dataset: {input_dataset}")
122
  dataset = load_dataset(input_dataset, split=split)
123
-
124
  # Validate image column
125
  if image_column not in dataset.column_names:
126
- raise ValueError(f"Column '{image_column}' not found. Available: {dataset.column_names}")
127
-
 
 
128
  # Limit samples if requested
129
  if max_samples:
130
  dataset = dataset.select(range(min(max_samples, len(dataset))))
131
  logger.info(f"Limited to {len(dataset)} samples")
132
-
133
  # Initialize vLLM
134
  logger.info(f"Initializing vLLM with model: {model}")
135
  llm = LLM(
@@ -139,53 +138,55 @@ def main(
139
  gpu_memory_utilization=gpu_memory_utilization,
140
  limit_mm_per_prompt={"image": 1},
141
  )
142
-
143
  sampling_params = SamplingParams(
144
  temperature=0.0, # Deterministic for OCR
145
  max_tokens=max_tokens,
146
  )
147
-
148
  # Process images in batches
149
  all_markdown = []
150
-
151
  logger.info(f"Processing {len(dataset)} images in batches of {batch_size}")
152
-
153
  # Process in batches to avoid memory issues
154
  for batch_indices in tqdm(
155
  partition_all(batch_size, range(len(dataset))),
156
  total=(len(dataset) + batch_size - 1) // batch_size,
157
- desc="OCR processing"
158
  ):
159
  batch_indices = list(batch_indices)
160
  batch_images = [dataset[i][image_column] for i in batch_indices]
161
-
162
  try:
163
  # Create messages for batch
164
  batch_messages = [make_ocr_message(img) for img in batch_images]
165
-
166
  # Process with vLLM
167
  outputs = llm.chat(batch_messages, sampling_params)
168
-
169
  # Extract markdown from outputs
170
  for output in outputs:
171
  markdown_text = output.outputs[0].text.strip()
172
  all_markdown.append(markdown_text)
173
-
174
  except Exception as e:
175
  logger.error(f"Error processing batch: {e}")
176
  # Add error placeholders for failed batch
177
  all_markdown.extend(["[OCR FAILED]"] * len(batch_images))
178
-
179
  # Add markdown column to dataset
180
  logger.info("Adding markdown column to dataset")
181
  dataset = dataset.add_column("markdown", all_markdown)
182
-
183
  # Push to hub
184
  logger.info(f"Pushing to {output_dataset}")
185
  dataset.push_to_hub(output_dataset, private=private, token=HF_TOKEN)
186
-
187
  logger.info("✅ OCR conversion complete!")
188
- logger.info(f"Dataset available at: https://huggingface.co/datasets/{output_dataset}")
 
 
189
 
190
 
191
  if __name__ == "__main__":
@@ -215,13 +216,15 @@ if __name__ == "__main__":
215
  print(" hfjobs run \\")
216
  print(" --flavor l4x1 \\")
217
  print(" --secret HF_TOKEN=... \\")
218
- print(" uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/nanonets-ocr.py \\")
 
 
219
  print(" your-document-dataset \\")
220
  print(" your-markdown-output")
221
  print("\n" + "=" * 80)
222
  print("\nFor full help, run: uv run nanonets-ocr.py --help")
223
  sys.exit(0)
224
-
225
  parser = argparse.ArgumentParser(
226
  description="OCR images to markdown using Nanonets-OCR-s",
227
  formatter_class=argparse.RawDescriptionHelpFormatter,
@@ -235,73 +238,60 @@ Examples:
235
 
236
  # Process subset for testing
237
  uv run nanonets-ocr.py large-dataset test-output --max-samples 100
238
- """
239
- )
240
-
241
- parser.add_argument(
242
- "input_dataset",
243
- help="Input dataset ID from Hugging Face Hub"
244
- )
245
- parser.add_argument(
246
- "output_dataset",
247
- help="Output dataset ID for Hugging Face Hub"
248
  )
 
 
 
249
  parser.add_argument(
250
  "--image-column",
251
  default="image",
252
- help="Column containing images (default: image)"
253
  )
254
  parser.add_argument(
255
  "--batch-size",
256
  type=int,
257
  default=8,
258
- help="Batch size for processing (default: 8)"
259
  )
260
  parser.add_argument(
261
  "--model",
262
  default="nanonets/Nanonets-OCR-s",
263
- help="Model to use (default: nanonets/Nanonets-OCR-s)"
264
  )
265
  parser.add_argument(
266
  "--max-model-len",
267
  type=int,
268
  default=8192,
269
- help="Maximum model context length (default: 8192)"
270
  )
271
  parser.add_argument(
272
  "--max-tokens",
273
  type=int,
274
  default=4096,
275
- help="Maximum tokens to generate (default: 4096)"
276
  )
277
  parser.add_argument(
278
  "--gpu-memory-utilization",
279
  type=float,
280
  default=0.7,
281
- help="GPU memory utilization (default: 0.7)"
282
- )
283
- parser.add_argument(
284
- "--hf-token",
285
- help="Hugging Face API token"
286
  )
 
287
  parser.add_argument(
288
- "--split",
289
- default="train",
290
- help="Dataset split to use (default: train)"
291
  )
292
  parser.add_argument(
293
  "--max-samples",
294
  type=int,
295
- help="Maximum number of samples to process (for testing)"
296
  )
297
  parser.add_argument(
298
- "--private",
299
- action="store_true",
300
- help="Make output dataset private"
301
  )
302
-
303
  args = parser.parse_args()
304
-
305
  main(
306
  input_dataset=args.input_dataset,
307
  output_dataset=args.output_dataset,
@@ -315,4 +305,4 @@ Examples:
315
  split=args.split,
316
  max_samples=args.max_samples,
317
  private=args.private,
318
- )
 
5
  # "huggingface-hub[hf_transfer]",
6
  # "pillow",
7
  # "vllm",
 
8
  # "tqdm",
9
  # "toolz",
10
  # "torch", # Added for CUDA check
11
  # ]
12
+ #
 
 
13
  # ///
14
 
15
  """
 
69
  pil_img = Image.open(image)
70
  else:
71
  raise ValueError(f"Unsupported image type: {type(image)}")
72
+
73
  # Convert to base64 data URI
74
  buf = io.BytesIO()
75
  pil_img.save(buf, format="PNG")
76
  data_uri = f"data:image/png;base64,{base64.b64encode(buf.getvalue()).decode()}"
77
+
78
  # Return message in vLLM format
79
  return [
80
  {
 
102
  private: bool = False,
103
  ):
104
  """Process images from HF dataset through OCR model."""
105
+
106
  # Check CUDA availability first
107
  check_cuda_availability()
108
+
109
  # Enable HF_TRANSFER for faster downloads
110
  os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
111
+
112
  # Login to HF if token provided
113
  HF_TOKEN = hf_token or os.environ.get("HF_TOKEN")
114
  if HF_TOKEN:
115
  login(token=HF_TOKEN)
116
+
117
  # Load dataset
118
  logger.info(f"Loading dataset: {input_dataset}")
119
  dataset = load_dataset(input_dataset, split=split)
120
+
121
  # Validate image column
122
  if image_column not in dataset.column_names:
123
+ raise ValueError(
124
+ f"Column '{image_column}' not found. Available: {dataset.column_names}"
125
+ )
126
+
127
  # Limit samples if requested
128
  if max_samples:
129
  dataset = dataset.select(range(min(max_samples, len(dataset))))
130
  logger.info(f"Limited to {len(dataset)} samples")
131
+
132
  # Initialize vLLM
133
  logger.info(f"Initializing vLLM with model: {model}")
134
  llm = LLM(
 
138
  gpu_memory_utilization=gpu_memory_utilization,
139
  limit_mm_per_prompt={"image": 1},
140
  )
141
+
142
  sampling_params = SamplingParams(
143
  temperature=0.0, # Deterministic for OCR
144
  max_tokens=max_tokens,
145
  )
146
+
147
  # Process images in batches
148
  all_markdown = []
149
+
150
  logger.info(f"Processing {len(dataset)} images in batches of {batch_size}")
151
+
152
  # Process in batches to avoid memory issues
153
  for batch_indices in tqdm(
154
  partition_all(batch_size, range(len(dataset))),
155
  total=(len(dataset) + batch_size - 1) // batch_size,
156
+ desc="OCR processing",
157
  ):
158
  batch_indices = list(batch_indices)
159
  batch_images = [dataset[i][image_column] for i in batch_indices]
160
+
161
  try:
162
  # Create messages for batch
163
  batch_messages = [make_ocr_message(img) for img in batch_images]
164
+
165
  # Process with vLLM
166
  outputs = llm.chat(batch_messages, sampling_params)
167
+
168
  # Extract markdown from outputs
169
  for output in outputs:
170
  markdown_text = output.outputs[0].text.strip()
171
  all_markdown.append(markdown_text)
172
+
173
  except Exception as e:
174
  logger.error(f"Error processing batch: {e}")
175
  # Add error placeholders for failed batch
176
  all_markdown.extend(["[OCR FAILED]"] * len(batch_images))
177
+
178
  # Add markdown column to dataset
179
  logger.info("Adding markdown column to dataset")
180
  dataset = dataset.add_column("markdown", all_markdown)
181
+
182
  # Push to hub
183
  logger.info(f"Pushing to {output_dataset}")
184
  dataset.push_to_hub(output_dataset, private=private, token=HF_TOKEN)
185
+
186
  logger.info("✅ OCR conversion complete!")
187
+ logger.info(
188
+ f"Dataset available at: https://huggingface.co/datasets/{output_dataset}"
189
+ )
190
 
191
 
192
  if __name__ == "__main__":
 
216
  print(" hfjobs run \\")
217
  print(" --flavor l4x1 \\")
218
  print(" --secret HF_TOKEN=... \\")
219
+ print(
220
+ " uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/nanonets-ocr.py \\"
221
+ )
222
  print(" your-document-dataset \\")
223
  print(" your-markdown-output")
224
  print("\n" + "=" * 80)
225
  print("\nFor full help, run: uv run nanonets-ocr.py --help")
226
  sys.exit(0)
227
+
228
  parser = argparse.ArgumentParser(
229
  description="OCR images to markdown using Nanonets-OCR-s",
230
  formatter_class=argparse.RawDescriptionHelpFormatter,
 
238
 
239
  # Process subset for testing
240
  uv run nanonets-ocr.py large-dataset test-output --max-samples 100
241
+ """,
 
 
 
 
 
 
 
 
 
242
  )
243
+
244
+ parser.add_argument("input_dataset", help="Input dataset ID from Hugging Face Hub")
245
+ parser.add_argument("output_dataset", help="Output dataset ID for Hugging Face Hub")
246
  parser.add_argument(
247
  "--image-column",
248
  default="image",
249
+ help="Column containing images (default: image)",
250
  )
251
  parser.add_argument(
252
  "--batch-size",
253
  type=int,
254
  default=8,
255
+ help="Batch size for processing (default: 8)",
256
  )
257
  parser.add_argument(
258
  "--model",
259
  default="nanonets/Nanonets-OCR-s",
260
+ help="Model to use (default: nanonets/Nanonets-OCR-s)",
261
  )
262
  parser.add_argument(
263
  "--max-model-len",
264
  type=int,
265
  default=8192,
266
+ help="Maximum model context length (default: 8192)",
267
  )
268
  parser.add_argument(
269
  "--max-tokens",
270
  type=int,
271
  default=4096,
272
+ help="Maximum tokens to generate (default: 4096)",
273
  )
274
  parser.add_argument(
275
  "--gpu-memory-utilization",
276
  type=float,
277
  default=0.7,
278
+ help="GPU memory utilization (default: 0.7)",
 
 
 
 
279
  )
280
+ parser.add_argument("--hf-token", help="Hugging Face API token")
281
  parser.add_argument(
282
+ "--split", default="train", help="Dataset split to use (default: train)"
 
 
283
  )
284
  parser.add_argument(
285
  "--max-samples",
286
  type=int,
287
+ help="Maximum number of samples to process (for testing)",
288
  )
289
  parser.add_argument(
290
+ "--private", action="store_true", help="Make output dataset private"
 
 
291
  )
292
+
293
  args = parser.parse_args()
294
+
295
  main(
296
  input_dataset=args.input_dataset,
297
  output_dataset=args.output_dataset,
 
305
  split=args.split,
306
  max_samples=args.max_samples,
307
  private=args.private,
308
+ )