simonJJJ commited on
Commit
8fc6465
1 Parent(s): 123dfa3
Files changed (1) hide show
  1. tokenization_qwen.py +5 -9
tokenization_qwen.py CHANGED
@@ -26,7 +26,7 @@ from matplotlib.font_manager import FontProperties
26
  logger = logging.getLogger(__name__)
27
 
28
 
29
- VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken"}
30
 
31
  PAT_STR = r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"""
32
  ENDOFTEXT = "<|endoftext|>"
@@ -169,9 +169,6 @@ class QWenTokenizer(PreTrainedTokenizer):
169
  self.im_start_id = self.special_tokens[IMSTART]
170
  self.im_end_id = self.special_tokens[IMEND]
171
 
172
- model_dir = getattr(self, 'model_dir', '')
173
- self.font_path = os.path.join(model_dir, "SimSun.ttf")
174
-
175
  def __len__(self) -> int:
176
  return self.tokenizer.n_vocab
177
 
@@ -417,8 +414,8 @@ class QWenTokenizer(PreTrainedTokenizer):
417
  h, w = image.height, image.width
418
  else:
419
  image = plt.imread(image)
420
- h, w = image.shape[0], image.shape[1]
421
- visualizer = Visualizer(image, font_path=self.font_path)
422
 
423
  boxes = self._fetch_all_box_with_ref(response)
424
  if not boxes:
@@ -493,9 +490,8 @@ class VisImage:
493
 
494
 
495
  class Visualizer:
496
- def __init__(self, img_rgb, font_path, metadata=None, scale=1.0):
497
  self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
498
- self.font_path = font_path
499
  self.output = VisImage(self.img, scale=scale)
500
  self.cpu_device = torch.device("cpu")
501
 
@@ -527,7 +523,7 @@ class Visualizer:
527
  y,
528
  text,
529
  size=font_size * self.output.scale,
530
- fontproperties=FontProperties(fname=self.font_path),
531
  bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
532
  verticalalignment="top",
533
  horizontalalignment=horizontal_alignment,
 
26
  logger = logging.getLogger(__name__)
27
 
28
 
29
+ VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken", "ttf": "SimSun.ttf"}
30
 
31
  PAT_STR = r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"""
32
  ENDOFTEXT = "<|endoftext|>"
 
169
  self.im_start_id = self.special_tokens[IMSTART]
170
  self.im_end_id = self.special_tokens[IMEND]
171
 
 
 
 
172
  def __len__(self) -> int:
173
  return self.tokenizer.n_vocab
174
 
 
414
  h, w = image.height, image.width
415
  else:
416
  image = plt.imread(image)
417
+ h, w = image.shape[0], image.shape[1]
418
+ visualizer = Visualizer(image)
419
 
420
  boxes = self._fetch_all_box_with_ref(response)
421
  if not boxes:
 
490
 
491
 
492
  class Visualizer:
493
+ def __init__(self, img_rgb, metadata=None, scale=1.0):
494
  self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
 
495
  self.output = VisImage(self.img, scale=scale)
496
  self.cpu_device = torch.device("cpu")
497
 
 
523
  y,
524
  text,
525
  size=font_size * self.output.scale,
526
+ fontproperties=FontProperties(fname=r"SimSun.ttf"),
527
  bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
528
  verticalalignment="top",
529
  horizontalalignment=horizontal_alignment,