Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
"""
|
| 2 |
OCR Application with Multiple Models including DeepSeek OCR
|
| 3 |
-
Fixed version with
|
| 4 |
"""
|
| 5 |
|
| 6 |
import os
|
|
@@ -97,7 +97,7 @@ except Exception as e:
|
|
| 97 |
processor_m = None
|
| 98 |
print(f"✗ olmOCR-2-7B-1025: Failed to load - {str(e)}")
|
| 99 |
|
| 100 |
-
# Load DeepSeek-OCR
|
| 101 |
try:
|
| 102 |
MODEL_ID_DS = "deepseek-ai/deepseek-ocr"
|
| 103 |
processor_ds = AutoProcessor.from_pretrained(MODEL_ID_DS, trust_remote_code=True)
|
|
@@ -106,7 +106,14 @@ try:
|
|
| 106 |
trust_remote_code=True,
|
| 107 |
torch_dtype=torch.float16
|
| 108 |
).eval()
|
| 109 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
except Exception as e:
|
| 111 |
model_ds = None
|
| 112 |
processor_ds = None
|
|
@@ -189,12 +196,17 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 189 |
]
|
| 190 |
}]
|
| 191 |
|
| 192 |
-
# Apply chat template
|
| 193 |
-
|
| 194 |
-
|
| 195 |
-
|
| 196 |
-
|
| 197 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 198 |
|
| 199 |
# Process inputs
|
| 200 |
inputs = processor(
|
|
@@ -239,6 +251,9 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 239 |
|
| 240 |
except Exception as e:
|
| 241 |
error_msg = f"Error during generation: {str(e)}"
|
|
|
|
|
|
|
|
|
|
| 242 |
yield error_msg, error_msg
|
| 243 |
|
| 244 |
|
|
|
|
| 1 |
"""
|
| 2 |
OCR Application with Multiple Models including DeepSeek OCR
|
| 3 |
+
Fixed version with proper chat template handling for DeepSeek OCR
|
| 4 |
"""
|
| 5 |
|
| 6 |
import os
|
|
|
|
| 97 |
processor_m = None
|
| 98 |
print(f"✗ olmOCR-2-7B-1025: Failed to load - {str(e)}")
|
| 99 |
|
| 100 |
+
# Load DeepSeek-OCR with custom chat template
|
| 101 |
try:
|
| 102 |
MODEL_ID_DS = "deepseek-ai/deepseek-ocr"
|
| 103 |
processor_ds = AutoProcessor.from_pretrained(MODEL_ID_DS, trust_remote_code=True)
|
|
|
|
| 106 |
trust_remote_code=True,
|
| 107 |
torch_dtype=torch.float16
|
| 108 |
).eval()
|
| 109 |
+
|
| 110 |
+
# Set a default chat template for DeepSeek OCR if it doesn't have one
|
| 111 |
+
if not hasattr(processor_ds.tokenizer, 'chat_template') or processor_ds.tokenizer.chat_template is None:
|
| 112 |
+
# Use a standard Qwen-style chat template
|
| 113 |
+
processor_ds.tokenizer.chat_template = "{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'text' %}{{ content['text'] }}{% endif %}{% endfor %}{% endif %}<|im_end|>\n{% endfor %}<|im_start|>assistant\n"
|
| 114 |
+
print("✓ DeepSeek-OCR loaded (with custom chat template)")
|
| 115 |
+
else:
|
| 116 |
+
print("✓ DeepSeek-OCR loaded")
|
| 117 |
except Exception as e:
|
| 118 |
model_ds = None
|
| 119 |
processor_ds = None
|
|
|
|
| 196 |
]
|
| 197 |
}]
|
| 198 |
|
| 199 |
+
# Apply chat template with error handling
|
| 200 |
+
try:
|
| 201 |
+
prompt_full = processor.apply_chat_template(
|
| 202 |
+
messages,
|
| 203 |
+
tokenize=False,
|
| 204 |
+
add_generation_prompt=True
|
| 205 |
+
)
|
| 206 |
+
except Exception as template_error:
|
| 207 |
+
# Fallback: create a simple prompt without chat template
|
| 208 |
+
print(f"Chat template error: {template_error}. Using fallback prompt.")
|
| 209 |
+
prompt_full = f"<|im_start|>user\n{text}<|im_end|>\n<|im_start|>assistant\n"
|
| 210 |
|
| 211 |
# Process inputs
|
| 212 |
inputs = processor(
|
|
|
|
| 251 |
|
| 252 |
except Exception as e:
|
| 253 |
error_msg = f"Error during generation: {str(e)}"
|
| 254 |
+
print(f"Full error: {e}")
|
| 255 |
+
import traceback
|
| 256 |
+
traceback.print_exc()
|
| 257 |
yield error_msg, error_msg
|
| 258 |
|
| 259 |
|