davanstrien HF Staff Claude commited on
Commit
e69f406
·
1 Parent(s): a9bc1eb

Fix output_path parameter in model.infer

Browse files

Add temp_output directory and pass real path instead of empty string:
- Create temp_output/ directory
- Pass to model.infer as output_path parameter
- Clean up both temp directories at end

The model expects a real directory path, not empty string.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Files changed (1) hide show
  1. deepseek-ocr.py +9 -3
deepseek-ocr.py CHANGED
@@ -201,6 +201,7 @@ def process_single_image(
201
  image_size: int,
202
  crop_mode: bool,
203
  temp_image_path: str,
 
204
  ) -> str:
205
  """Process a single image through DeepSeek-OCR."""
206
  # Convert to RGB if needed
@@ -215,7 +216,7 @@ def process_single_image(
215
  tokenizer,
216
  prompt=prompt,
217
  image_file=temp_image_path,
218
- output_path="", # Don't save intermediate files
219
  base_size=base_size,
220
  image_size=image_size,
221
  crop_mode=crop_mode,
@@ -333,11 +334,14 @@ def main(
333
  logger.info(f"Processing {len(dataset)} images (sequential, no batching)")
334
  logger.info("Note: This may be slower than vLLM-based scripts")
335
 
336
- # Create temp directory for image files (simple local dir)
337
  temp_dir = Path("temp_images")
338
  temp_dir.mkdir(exist_ok=True)
339
  temp_image_path = str(temp_dir / "temp_image.png")
340
 
 
 
 
341
  try:
342
  for i in tqdm(range(len(dataset)), desc="OCR processing"):
343
  try:
@@ -362,6 +366,7 @@ def main(
362
  final_image_size,
363
  final_crop_mode,
364
  temp_image_path,
 
365
  )
366
 
367
  all_markdown.append(result)
@@ -371,9 +376,10 @@ def main(
371
  all_markdown.append("[OCR FAILED]")
372
 
373
  finally:
374
- # Clean up temp directory
375
  try:
376
  shutil.rmtree(temp_dir)
 
377
  except:
378
  pass
379
 
 
201
  image_size: int,
202
  crop_mode: bool,
203
  temp_image_path: str,
204
+ temp_output_dir: str,
205
  ) -> str:
206
  """Process a single image through DeepSeek-OCR."""
207
  # Convert to RGB if needed
 
216
  tokenizer,
217
  prompt=prompt,
218
  image_file=temp_image_path,
219
+ output_path=temp_output_dir, # Need real directory path
220
  base_size=base_size,
221
  image_size=image_size,
222
  crop_mode=crop_mode,
 
334
  logger.info(f"Processing {len(dataset)} images (sequential, no batching)")
335
  logger.info("Note: This may be slower than vLLM-based scripts")
336
 
337
+ # Create temp directories for image files and output (simple local dirs)
338
  temp_dir = Path("temp_images")
339
  temp_dir.mkdir(exist_ok=True)
340
  temp_image_path = str(temp_dir / "temp_image.png")
341
 
342
+ temp_output_dir = Path("temp_output")
343
+ temp_output_dir.mkdir(exist_ok=True)
344
+
345
  try:
346
  for i in tqdm(range(len(dataset)), desc="OCR processing"):
347
  try:
 
366
  final_image_size,
367
  final_crop_mode,
368
  temp_image_path,
369
+ str(temp_output_dir),
370
  )
371
 
372
  all_markdown.append(result)
 
376
  all_markdown.append("[OCR FAILED]")
377
 
378
  finally:
379
+ # Clean up temp directories
380
  try:
381
  shutil.rmtree(temp_dir)
382
+ shutil.rmtree(temp_output_dir)
383
  except:
384
  pass
385