Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,3 @@
|
|
| 1 |
-
|
| 2 |
import os
|
| 3 |
import time
|
| 4 |
import torch
|
|
@@ -18,13 +17,18 @@ from transformers import (
|
|
| 18 |
Qwen2_5_VLForConditionalGeneration,
|
| 19 |
TextIteratorStreamer
|
| 20 |
)
|
|
|
|
| 21 |
from qwen_vl_utils import process_vision_info
|
| 22 |
|
| 23 |
|
|
|
|
|
|
|
| 24 |
# Suppress the warning about uninitialized weights
|
| 25 |
warnings.filterwarnings('ignore', message='Some weights.*were not initialized')
|
| 26 |
|
| 27 |
|
|
|
|
|
|
|
| 28 |
# Try importing Qwen3VL if available
|
| 29 |
try:
|
| 30 |
from transformers import Qwen3VLForConditionalGeneration
|
|
@@ -32,18 +36,27 @@ except ImportError:
|
|
| 32 |
Qwen3VLForConditionalGeneration = None
|
| 33 |
|
| 34 |
|
|
|
|
|
|
|
| 35 |
MAX_MAX_NEW_TOKENS = 4096
|
| 36 |
DEFAULT_MAX_NEW_TOKENS = 2048
|
| 37 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
|
|
|
|
|
|
|
|
|
| 38 |
|
| 39 |
|
| 40 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
| 41 |
|
| 42 |
|
|
|
|
|
|
|
| 43 |
print(f"Initial Device: {device}")
|
| 44 |
print(f"CUDA Available: {torch.cuda.is_available()}")
|
| 45 |
|
| 46 |
|
|
|
|
|
|
|
| 47 |
# Load Chandra-OCR
|
| 48 |
try:
|
| 49 |
MODEL_ID_V = "datalab-to/chandra"
|
|
@@ -52,7 +65,8 @@ try:
|
|
| 52 |
model_v = Qwen3VLForConditionalGeneration.from_pretrained(
|
| 53 |
MODEL_ID_V,
|
| 54 |
trust_remote_code=True,
|
| 55 |
-
torch_dtype=torch.float16
|
|
|
|
| 56 |
).eval()
|
| 57 |
print("✓ Chandra-OCR loaded")
|
| 58 |
else:
|
|
@@ -64,6 +78,8 @@ except Exception as e:
|
|
| 64 |
print(f"✗ Chandra-OCR: Failed to load - {str(e)}")
|
| 65 |
|
| 66 |
|
|
|
|
|
|
|
| 67 |
# Load Nanonets-OCR2-3B
|
| 68 |
try:
|
| 69 |
MODEL_ID_X = "nanonets/Nanonets-OCR2-3B"
|
|
@@ -71,7 +87,8 @@ try:
|
|
| 71 |
model_x = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 72 |
MODEL_ID_X,
|
| 73 |
trust_remote_code=True,
|
| 74 |
-
torch_dtype=torch.float16
|
|
|
|
| 75 |
).eval()
|
| 76 |
print("✓ Nanonets-OCR2-3B loaded")
|
| 77 |
except Exception as e:
|
|
@@ -80,14 +97,31 @@ except Exception as e:
|
|
| 80 |
print(f"✗ Nanonets-OCR2-3B: Failed to load - {str(e)}")
|
| 81 |
|
| 82 |
|
| 83 |
-
|
|
|
|
|
|
|
| 84 |
try:
|
| 85 |
-
|
| 86 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
model_d = AutoModelForCausalLM.from_pretrained(
|
| 88 |
-
|
| 89 |
attn_implementation="flash_attention_2",
|
| 90 |
torch_dtype=torch.bfloat16,
|
|
|
|
| 91 |
trust_remote_code=True
|
| 92 |
).eval()
|
| 93 |
print("✓ Dots.OCR loaded")
|
|
@@ -95,6 +129,10 @@ except Exception as e:
|
|
| 95 |
model_d = None
|
| 96 |
processor_d = None
|
| 97 |
print(f"✗ Dots.OCR: Failed to load - {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 98 |
|
| 99 |
|
| 100 |
# Load olmOCR-2-7B-1025
|
|
@@ -104,7 +142,8 @@ try:
|
|
| 104 |
model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 105 |
MODEL_ID_M,
|
| 106 |
trust_remote_code=True,
|
| 107 |
-
torch_dtype=torch.float16
|
|
|
|
| 108 |
).eval()
|
| 109 |
print("✓ olmOCR-2-7B-1025 loaded")
|
| 110 |
except Exception as e:
|
|
@@ -113,6 +152,8 @@ except Exception as e:
|
|
| 113 |
print(f"✗ olmOCR-2-7B-1025: Failed to load - {str(e)}")
|
| 114 |
|
| 115 |
|
|
|
|
|
|
|
| 116 |
@spaces.GPU
|
| 117 |
def generate_image(model_name: str, text: str, image: Image.Image,
|
| 118 |
max_new_tokens: int, temperature: float, top_p: float,
|
|
@@ -120,10 +161,8 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 120 |
"""
|
| 121 |
Generates responses using the selected model for image input.
|
| 122 |
Yields raw text and Markdown-formatted text.
|
| 123 |
-
|
| 124 |
This function is decorated with @spaces.GPU to ensure it runs on GPU
|
| 125 |
when available in Hugging Face Spaces.
|
| 126 |
-
|
| 127 |
Args:
|
| 128 |
model_name: Name of the OCR model to use
|
| 129 |
text: Prompt text for the model
|
|
@@ -133,48 +172,52 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 133 |
top_p: Nucleus sampling parameter
|
| 134 |
top_k: Top-k sampling parameter
|
| 135 |
repetition_penalty: Penalty for repeating tokens
|
| 136 |
-
|
| 137 |
Yields:
|
| 138 |
tuple: (raw_text, markdown_text)
|
| 139 |
"""
|
| 140 |
# Device will be cuda when @spaces.GPU decorator activates
|
| 141 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
| 142 |
|
|
|
|
| 143 |
# Select model and processor based on model_name
|
| 144 |
if model_name == "olmOCR-2-7B-1025":
|
| 145 |
if model_m is None:
|
| 146 |
yield "olmOCR-2-7B-1025 is not available.", "olmOCR-2-7B-1025 is not available."
|
| 147 |
return
|
| 148 |
processor = processor_m
|
| 149 |
-
model = model_m
|
| 150 |
elif model_name == "Nanonets-OCR2-3B":
|
| 151 |
if model_x is None:
|
| 152 |
yield "Nanonets-OCR2-3B is not available.", "Nanonets-OCR2-3B is not available."
|
| 153 |
return
|
| 154 |
processor = processor_x
|
| 155 |
-
model = model_x
|
| 156 |
elif model_name == "Chandra-OCR":
|
| 157 |
if model_v is None:
|
| 158 |
yield "Chandra-OCR is not available.", "Chandra-OCR is not available."
|
| 159 |
return
|
| 160 |
processor = processor_v
|
| 161 |
-
model = model_v
|
| 162 |
elif model_name == "Dots.OCR":
|
| 163 |
if model_d is None:
|
| 164 |
yield "Dots.OCR is not available.", "Dots.OCR is not available."
|
| 165 |
return
|
| 166 |
processor = processor_d
|
| 167 |
-
model = model_d
|
| 168 |
else:
|
| 169 |
yield "Invalid model selected.", "Invalid model selected."
|
| 170 |
return
|
| 171 |
|
| 172 |
|
|
|
|
|
|
|
| 173 |
if image is None:
|
| 174 |
yield "Please upload an image.", "Please upload an image."
|
| 175 |
return
|
| 176 |
|
| 177 |
|
|
|
|
|
|
|
| 178 |
try:
|
| 179 |
# Prepare messages in chat format
|
| 180 |
messages = [{
|
|
@@ -185,6 +228,7 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 185 |
]
|
| 186 |
}]
|
| 187 |
|
|
|
|
| 188 |
# Apply chat template with fallback
|
| 189 |
try:
|
| 190 |
prompt_full = processor.apply_chat_template(
|
|
@@ -198,6 +242,8 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 198 |
prompt_full = f"{text}"
|
| 199 |
|
| 200 |
|
|
|
|
|
|
|
| 201 |
# Process inputs
|
| 202 |
inputs = processor(
|
| 203 |
text=[prompt_full],
|
|
@@ -207,6 +253,8 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 207 |
).to(device)
|
| 208 |
|
| 209 |
|
|
|
|
|
|
|
| 210 |
# Setup streaming generation
|
| 211 |
streamer = TextIteratorStreamer(
|
| 212 |
processor.tokenizer if hasattr(processor, 'tokenizer') else processor,
|
|
@@ -214,6 +262,7 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 214 |
skip_special_tokens=True
|
| 215 |
)
|
| 216 |
|
|
|
|
| 217 |
generation_kwargs = {
|
| 218 |
**inputs,
|
| 219 |
"streamer": streamer,
|
|
@@ -225,10 +274,12 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 225 |
"repetition_penalty": repetition_penalty,
|
| 226 |
}
|
| 227 |
|
|
|
|
| 228 |
# Start generation in separate thread
|
| 229 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
| 230 |
thread.start()
|
| 231 |
|
|
|
|
| 232 |
# Stream the results
|
| 233 |
buffer = ""
|
| 234 |
for new_text in streamer:
|
|
@@ -237,9 +288,11 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 237 |
time.sleep(0.01)
|
| 238 |
yield buffer, buffer
|
| 239 |
|
|
|
|
| 240 |
# Ensure thread completes
|
| 241 |
thread.join()
|
| 242 |
|
|
|
|
| 243 |
except Exception as e:
|
| 244 |
error_msg = f"Error during generation: {str(e)}"
|
| 245 |
print(f"Full error: {e}")
|
|
@@ -248,10 +301,13 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
| 248 |
yield error_msg, error_msg
|
| 249 |
|
| 250 |
|
|
|
|
|
|
|
| 251 |
# Example usage for Gradio interface
|
| 252 |
if __name__ == "__main__":
|
| 253 |
import gradio as gr
|
| 254 |
|
|
|
|
| 255 |
# Determine available models
|
| 256 |
available_models = []
|
| 257 |
if model_m is not None:
|
|
@@ -267,16 +323,20 @@ if __name__ == "__main__":
|
|
| 267 |
available_models.append("Dots.OCR")
|
| 268 |
print(" Added: Dots.OCR")
|
| 269 |
|
|
|
|
| 270 |
if not available_models:
|
| 271 |
print("ERROR: No models were loaded successfully!")
|
| 272 |
exit(1)
|
| 273 |
|
|
|
|
| 274 |
print(f"\n✓ Available models for dropdown: {', '.join(available_models)}")
|
| 275 |
|
|
|
|
| 276 |
with gr.Blocks(title="Multi-Model OCR") as demo:
|
| 277 |
gr.Markdown("# 🔍 Multi-Model OCR Application")
|
| 278 |
gr.Markdown("Upload an image and select a model to extract text. Models run on GPU via Hugging Face Spaces.")
|
| 279 |
|
|
|
|
| 280 |
with gr.Row():
|
| 281 |
with gr.Column():
|
| 282 |
model_selector = gr.Dropdown(
|
|
@@ -291,6 +351,7 @@ if __name__ == "__main__":
|
|
| 291 |
lines=2
|
| 292 |
)
|
| 293 |
|
|
|
|
| 294 |
with gr.Accordion("Advanced Settings", open=False):
|
| 295 |
max_tokens = gr.Slider(
|
| 296 |
minimum=1,
|
|
@@ -328,20 +389,24 @@ if __name__ == "__main__":
|
|
| 328 |
label="Repetition Penalty"
|
| 329 |
)
|
| 330 |
|
|
|
|
| 331 |
submit_btn = gr.Button("Extract Text", variant="primary")
|
| 332 |
|
|
|
|
| 333 |
with gr.Column():
|
| 334 |
output_text = gr.Textbox(label="Extracted Text", lines=20)
|
| 335 |
output_markdown = gr.Markdown(label="Formatted Output")
|
| 336 |
|
|
|
|
| 337 |
gr.Markdown("""
|
| 338 |
### Available Models:
|
| 339 |
- **olmOCR-2-7B-1025**: Allen AI's OCR model
|
| 340 |
- **Nanonets-OCR2-3B**: Nanonets OCR model
|
| 341 |
- **Chandra-OCR**: Datalab OCR model
|
| 342 |
-
- **Dots.OCR**: Stranger Vision OCR model
|
| 343 |
""")
|
| 344 |
|
|
|
|
| 345 |
submit_btn.click(
|
| 346 |
fn=generate_image,
|
| 347 |
inputs=[
|
|
@@ -357,5 +422,6 @@ if __name__ == "__main__":
|
|
| 357 |
outputs=[output_text, output_markdown]
|
| 358 |
)
|
| 359 |
|
|
|
|
| 360 |
# Launch with share=True for Hugging Face Spaces
|
| 361 |
-
demo.launch(share=True)
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
import time
|
| 3 |
import torch
|
|
|
|
| 17 |
Qwen2_5_VLForConditionalGeneration,
|
| 18 |
TextIteratorStreamer
|
| 19 |
)
|
| 20 |
+
from huggingface_hub import snapshot_download
|
| 21 |
from qwen_vl_utils import process_vision_info
|
| 22 |
|
| 23 |
|
| 24 |
+
|
| 25 |
+
|
| 26 |
# Suppress the warning about uninitialized weights
|
| 27 |
warnings.filterwarnings('ignore', message='Some weights.*were not initialized')
|
| 28 |
|
| 29 |
|
| 30 |
+
|
| 31 |
+
|
| 32 |
# Try importing Qwen3VL if available
|
| 33 |
try:
|
| 34 |
from transformers import Qwen3VLForConditionalGeneration
|
|
|
|
| 36 |
Qwen3VLForConditionalGeneration = None
|
| 37 |
|
| 38 |
|
| 39 |
+
|
| 40 |
+
|
| 41 |
MAX_MAX_NEW_TOKENS = 4096
|
| 42 |
DEFAULT_MAX_NEW_TOKENS = 2048
|
| 43 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
| 44 |
+
CACHE_DIR = os.getenv("HF_CACHE_DIR", "./models")
|
| 45 |
+
|
| 46 |
+
|
| 47 |
|
| 48 |
|
| 49 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
| 50 |
|
| 51 |
|
| 52 |
+
|
| 53 |
+
|
| 54 |
print(f"Initial Device: {device}")
|
| 55 |
print(f"CUDA Available: {torch.cuda.is_available()}")
|
| 56 |
|
| 57 |
|
| 58 |
+
|
| 59 |
+
|
| 60 |
# Load Chandra-OCR
|
| 61 |
try:
|
| 62 |
MODEL_ID_V = "datalab-to/chandra"
|
|
|
|
| 65 |
model_v = Qwen3VLForConditionalGeneration.from_pretrained(
|
| 66 |
MODEL_ID_V,
|
| 67 |
trust_remote_code=True,
|
| 68 |
+
torch_dtype=torch.float16,
|
| 69 |
+
device_map="auto"
|
| 70 |
).eval()
|
| 71 |
print("✓ Chandra-OCR loaded")
|
| 72 |
else:
|
|
|
|
| 78 |
print(f"✗ Chandra-OCR: Failed to load - {str(e)}")
|
| 79 |
|
| 80 |
|
| 81 |
+
|
| 82 |
+
|
| 83 |
# Load Nanonets-OCR2-3B
|
| 84 |
try:
|
| 85 |
MODEL_ID_X = "nanonets/Nanonets-OCR2-3B"
|
|
|
|
| 87 |
model_x = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 88 |
MODEL_ID_X,
|
| 89 |
trust_remote_code=True,
|
| 90 |
+
torch_dtype=torch.float16,
|
| 91 |
+
device_map="auto"
|
| 92 |
).eval()
|
| 93 |
print("✓ Nanonets-OCR2-3B loaded")
|
| 94 |
except Exception as e:
|
|
|
|
| 97 |
print(f"✗ Nanonets-OCR2-3B: Failed to load - {str(e)}")
|
| 98 |
|
| 99 |
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
# Load Dots.OCR - UPDATED with snapshot_download and device_map="auto"
|
| 103 |
try:
|
| 104 |
+
MODEL_ID_D = "rednote-hilab/dots.ocr"
|
| 105 |
+
model_path_d = os.path.join(CACHE_DIR, "dots-ocr-local")
|
| 106 |
+
|
| 107 |
+
# Download and cache model locally
|
| 108 |
+
snapshot_download(
|
| 109 |
+
repo_id=MODEL_ID_D,
|
| 110 |
+
local_dir=model_path_d,
|
| 111 |
+
local_dir_use_symlinks=False, # Avoid symlink issues on HF Spaces
|
| 112 |
+
allow_patterns=["*.json", "*.bin", "*.safetensors", "*.txt"]
|
| 113 |
+
)
|
| 114 |
+
|
| 115 |
+
processor_d = AutoProcessor.from_pretrained(
|
| 116 |
+
model_path_d,
|
| 117 |
+
trust_remote_code=True
|
| 118 |
+
)
|
| 119 |
+
|
| 120 |
model_d = AutoModelForCausalLM.from_pretrained(
|
| 121 |
+
model_path_d,
|
| 122 |
attn_implementation="flash_attention_2",
|
| 123 |
torch_dtype=torch.bfloat16,
|
| 124 |
+
device_map="auto", # Better memory management
|
| 125 |
trust_remote_code=True
|
| 126 |
).eval()
|
| 127 |
print("✓ Dots.OCR loaded")
|
|
|
|
| 129 |
model_d = None
|
| 130 |
processor_d = None
|
| 131 |
print(f"✗ Dots.OCR: Failed to load - {str(e)}")
|
| 132 |
+
import traceback
|
| 133 |
+
traceback.print_exc()
|
| 134 |
+
|
| 135 |
+
|
| 136 |
|
| 137 |
|
| 138 |
# Load olmOCR-2-7B-1025
|
|
|
|
| 142 |
model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
| 143 |
MODEL_ID_M,
|
| 144 |
trust_remote_code=True,
|
| 145 |
+
torch_dtype=torch.float16,
|
| 146 |
+
device_map="auto"
|
| 147 |
).eval()
|
| 148 |
print("✓ olmOCR-2-7B-1025 loaded")
|
| 149 |
except Exception as e:
|
|
|
|
| 152 |
print(f"✗ olmOCR-2-7B-1025: Failed to load - {str(e)}")
|
| 153 |
|
| 154 |
|
| 155 |
+
|
| 156 |
+
|
| 157 |
@spaces.GPU
|
| 158 |
def generate_image(model_name: str, text: str, image: Image.Image,
|
| 159 |
max_new_tokens: int, temperature: float, top_p: float,
|
|
|
|
| 161 |
"""
|
| 162 |
Generates responses using the selected model for image input.
|
| 163 |
Yields raw text and Markdown-formatted text.
|
|
|
|
| 164 |
This function is decorated with @spaces.GPU to ensure it runs on GPU
|
| 165 |
when available in Hugging Face Spaces.
|
|
|
|
| 166 |
Args:
|
| 167 |
model_name: Name of the OCR model to use
|
| 168 |
text: Prompt text for the model
|
|
|
|
| 172 |
top_p: Nucleus sampling parameter
|
| 173 |
top_k: Top-k sampling parameter
|
| 174 |
repetition_penalty: Penalty for repeating tokens
|
|
|
|
| 175 |
Yields:
|
| 176 |
tuple: (raw_text, markdown_text)
|
| 177 |
"""
|
| 178 |
# Device will be cuda when @spaces.GPU decorator activates
|
| 179 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
| 180 |
|
| 181 |
+
|
| 182 |
# Select model and processor based on model_name
|
| 183 |
if model_name == "olmOCR-2-7B-1025":
|
| 184 |
if model_m is None:
|
| 185 |
yield "olmOCR-2-7B-1025 is not available.", "olmOCR-2-7B-1025 is not available."
|
| 186 |
return
|
| 187 |
processor = processor_m
|
| 188 |
+
model = model_m
|
| 189 |
elif model_name == "Nanonets-OCR2-3B":
|
| 190 |
if model_x is None:
|
| 191 |
yield "Nanonets-OCR2-3B is not available.", "Nanonets-OCR2-3B is not available."
|
| 192 |
return
|
| 193 |
processor = processor_x
|
| 194 |
+
model = model_x
|
| 195 |
elif model_name == "Chandra-OCR":
|
| 196 |
if model_v is None:
|
| 197 |
yield "Chandra-OCR is not available.", "Chandra-OCR is not available."
|
| 198 |
return
|
| 199 |
processor = processor_v
|
| 200 |
+
model = model_v
|
| 201 |
elif model_name == "Dots.OCR":
|
| 202 |
if model_d is None:
|
| 203 |
yield "Dots.OCR is not available.", "Dots.OCR is not available."
|
| 204 |
return
|
| 205 |
processor = processor_d
|
| 206 |
+
model = model_d
|
| 207 |
else:
|
| 208 |
yield "Invalid model selected.", "Invalid model selected."
|
| 209 |
return
|
| 210 |
|
| 211 |
|
| 212 |
+
|
| 213 |
+
|
| 214 |
if image is None:
|
| 215 |
yield "Please upload an image.", "Please upload an image."
|
| 216 |
return
|
| 217 |
|
| 218 |
|
| 219 |
+
|
| 220 |
+
|
| 221 |
try:
|
| 222 |
# Prepare messages in chat format
|
| 223 |
messages = [{
|
|
|
|
| 228 |
]
|
| 229 |
}]
|
| 230 |
|
| 231 |
+
|
| 232 |
# Apply chat template with fallback
|
| 233 |
try:
|
| 234 |
prompt_full = processor.apply_chat_template(
|
|
|
|
| 242 |
prompt_full = f"{text}"
|
| 243 |
|
| 244 |
|
| 245 |
+
|
| 246 |
+
|
| 247 |
# Process inputs
|
| 248 |
inputs = processor(
|
| 249 |
text=[prompt_full],
|
|
|
|
| 253 |
).to(device)
|
| 254 |
|
| 255 |
|
| 256 |
+
|
| 257 |
+
|
| 258 |
# Setup streaming generation
|
| 259 |
streamer = TextIteratorStreamer(
|
| 260 |
processor.tokenizer if hasattr(processor, 'tokenizer') else processor,
|
|
|
|
| 262 |
skip_special_tokens=True
|
| 263 |
)
|
| 264 |
|
| 265 |
+
|
| 266 |
generation_kwargs = {
|
| 267 |
**inputs,
|
| 268 |
"streamer": streamer,
|
|
|
|
| 274 |
"repetition_penalty": repetition_penalty,
|
| 275 |
}
|
| 276 |
|
| 277 |
+
|
| 278 |
# Start generation in separate thread
|
| 279 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
| 280 |
thread.start()
|
| 281 |
|
| 282 |
+
|
| 283 |
# Stream the results
|
| 284 |
buffer = ""
|
| 285 |
for new_text in streamer:
|
|
|
|
| 288 |
time.sleep(0.01)
|
| 289 |
yield buffer, buffer
|
| 290 |
|
| 291 |
+
|
| 292 |
# Ensure thread completes
|
| 293 |
thread.join()
|
| 294 |
|
| 295 |
+
|
| 296 |
except Exception as e:
|
| 297 |
error_msg = f"Error during generation: {str(e)}"
|
| 298 |
print(f"Full error: {e}")
|
|
|
|
| 301 |
yield error_msg, error_msg
|
| 302 |
|
| 303 |
|
| 304 |
+
|
| 305 |
+
|
| 306 |
# Example usage for Gradio interface
|
| 307 |
if __name__ == "__main__":
|
| 308 |
import gradio as gr
|
| 309 |
|
| 310 |
+
|
| 311 |
# Determine available models
|
| 312 |
available_models = []
|
| 313 |
if model_m is not None:
|
|
|
|
| 323 |
available_models.append("Dots.OCR")
|
| 324 |
print(" Added: Dots.OCR")
|
| 325 |
|
| 326 |
+
|
| 327 |
if not available_models:
|
| 328 |
print("ERROR: No models were loaded successfully!")
|
| 329 |
exit(1)
|
| 330 |
|
| 331 |
+
|
| 332 |
print(f"\n✓ Available models for dropdown: {', '.join(available_models)}")
|
| 333 |
|
| 334 |
+
|
| 335 |
with gr.Blocks(title="Multi-Model OCR") as demo:
|
| 336 |
gr.Markdown("# 🔍 Multi-Model OCR Application")
|
| 337 |
gr.Markdown("Upload an image and select a model to extract text. Models run on GPU via Hugging Face Spaces.")
|
| 338 |
|
| 339 |
+
|
| 340 |
with gr.Row():
|
| 341 |
with gr.Column():
|
| 342 |
model_selector = gr.Dropdown(
|
|
|
|
| 351 |
lines=2
|
| 352 |
)
|
| 353 |
|
| 354 |
+
|
| 355 |
with gr.Accordion("Advanced Settings", open=False):
|
| 356 |
max_tokens = gr.Slider(
|
| 357 |
minimum=1,
|
|
|
|
| 389 |
label="Repetition Penalty"
|
| 390 |
)
|
| 391 |
|
| 392 |
+
|
| 393 |
submit_btn = gr.Button("Extract Text", variant="primary")
|
| 394 |
|
| 395 |
+
|
| 396 |
with gr.Column():
|
| 397 |
output_text = gr.Textbox(label="Extracted Text", lines=20)
|
| 398 |
output_markdown = gr.Markdown(label="Formatted Output")
|
| 399 |
|
| 400 |
+
|
| 401 |
gr.Markdown("""
|
| 402 |
### Available Models:
|
| 403 |
- **olmOCR-2-7B-1025**: Allen AI's OCR model
|
| 404 |
- **Nanonets-OCR2-3B**: Nanonets OCR model
|
| 405 |
- **Chandra-OCR**: Datalab OCR model
|
| 406 |
+
- **Dots.OCR**: Stranger Vision OCR model (Updated)
|
| 407 |
""")
|
| 408 |
|
| 409 |
+
|
| 410 |
submit_btn.click(
|
| 411 |
fn=generate_image,
|
| 412 |
inputs=[
|
|
|
|
| 422 |
outputs=[output_text, output_markdown]
|
| 423 |
)
|
| 424 |
|
| 425 |
+
|
| 426 |
# Launch with share=True for Hugging Face Spaces
|
| 427 |
+
demo.launch(share=True)
|