Spaces:
Running
on
Zero
Running
on
Zero
Fix app.py
Browse files
app.py
CHANGED
@@ -2,7 +2,7 @@ import os
|
|
2 |
import warnings
|
3 |
import torch
|
4 |
import gc
|
5 |
-
from transformers import AutoModelForVision2Seq, AutoProcessor
|
6 |
from PIL import Image
|
7 |
import gradio as gr
|
8 |
from huggingface_hub import login
|
@@ -17,115 +17,106 @@ processor = None
|
|
17 |
|
18 |
# Clear CUDA cache
|
19 |
if torch.cuda.is_available():
|
20 |
-
|
21 |
-
|
22 |
print("เคลียร์ CUDA cache เรียบร้อยแล้ว")
|
23 |
|
24 |
# Login to Hugging Face Hub
|
25 |
if 'HUGGING_FACE_HUB_TOKEN' in os.environ:
|
26 |
-
|
27 |
-
|
28 |
else:
|
29 |
-
|
30 |
|
31 |
def load_model_and_processor():
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
bnb_4bit_use_double_quant=True,
|
44 |
-
bnb_4bit_quant_type="nf4",
|
45 |
-
bnb_4bit_compute_dtype=torch.bfloat16
|
46 |
-
)
|
47 |
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
print("โหลดโมเดลสำเร็จ!")
|
63 |
-
return True
|
64 |
-
except Exception as e:
|
65 |
-
print(f"เกิดข้อผิดพลาดในการโหลดโมเดล: {str(e)}")
|
66 |
-
return False
|
67 |
|
68 |
def process_handwriting(image):
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
Only return the transcription in Thai language."""
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
|
115 |
# Initialize application
|
116 |
print("กำลังเริ่มต้นแอปพลิเคชัน...")
|
117 |
if load_model_and_processor():
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
else:
|
131 |
-
|
|
|
2 |
import warnings
|
3 |
import torch
|
4 |
import gc
|
5 |
+
from transformers import AutoModelForVision2Seq, AutoProcessor
|
6 |
from PIL import Image
|
7 |
import gradio as gr
|
8 |
from huggingface_hub import login
|
|
|
17 |
|
18 |
# Clear CUDA cache
|
19 |
if torch.cuda.is_available():
|
20 |
+
torch.cuda.empty_cache()
|
21 |
+
gc.collect()
|
22 |
print("เคลียร์ CUDA cache เรียบร้อยแล้ว")
|
23 |
|
24 |
# Login to Hugging Face Hub
|
25 |
if 'HUGGING_FACE_HUB_TOKEN' in os.environ:
|
26 |
+
print("กำลังเข้าสู่ระบบ Hugging Face Hub...")
|
27 |
+
login(token=os.environ['HUGGING_FACE_HUB_TOKEN'])
|
28 |
else:
|
29 |
+
print("คำเตือน: ไม่พบ HUGGING_FACE_HUB_TOKEN")
|
30 |
|
31 |
def load_model_and_processor():
|
32 |
+
"""โหลดโมเดลและ processor"""
|
33 |
+
global model, processor
|
34 |
+
print("กำลังโหลดโมเดลและ processor...")
|
35 |
+
try:
|
36 |
+
# Model paths
|
37 |
+
base_model_path = "meta-llama/Llama-3.2-11B-Vision-Instruct"
|
38 |
+
hub_model_path = "Aekanun/thai-handwriting-llm"
|
39 |
|
40 |
+
# Load processor from base model
|
41 |
+
print("กำลังโหลด processor...")
|
42 |
+
processor = AutoProcessor.from_pretrained(base_model_path, use_auth_token=True)
|
|
|
|
|
|
|
|
|
43 |
|
44 |
+
# Load model from Hub
|
45 |
+
print("กำลังโหลดโมเดลจาก Hub...")
|
46 |
+
model = AutoModelForVision2Seq.from_pretrained(
|
47 |
+
hub_model_path,
|
48 |
+
device_map="auto",
|
49 |
+
torch_dtype=torch.bfloat16,
|
50 |
+
trust_remote_code=True,
|
51 |
+
use_auth_token=True
|
52 |
+
)
|
53 |
+
print("โหลดโมเดลสำเร็จ!")
|
54 |
+
return True
|
55 |
+
except Exception as e:
|
56 |
+
print(f"เกิดข้อผิดพลาดในการโหลดโมเดล: {str(e)}")
|
57 |
+
return False
|
|
|
|
|
|
|
|
|
|
|
58 |
|
59 |
def process_handwriting(image):
|
60 |
+
"""ฟังก์ชันสำหรับ Gradio interface"""
|
61 |
+
global model, processor
|
62 |
+
|
63 |
+
if image is None:
|
64 |
+
return "กรุณาอัพโหลดรูปภาพ"
|
65 |
+
|
66 |
+
try:
|
67 |
+
# Ensure image is in PIL format
|
68 |
+
if not isinstance(image, Image.Image):
|
69 |
+
image = Image.fromarray(image)
|
70 |
+
|
71 |
+
# Create prompt
|
72 |
+
prompt = """Transcribe the Thai handwritten text from the provided image.
|
73 |
Only return the transcription in Thai language."""
|
74 |
+
|
75 |
+
# Create model inputs
|
76 |
+
messages = [
|
77 |
+
{
|
78 |
+
"role": "user",
|
79 |
+
"content": [
|
80 |
+
{"type": "text", "text": prompt},
|
81 |
+
{"type": "image", "image": image}
|
82 |
+
],
|
83 |
+
}
|
84 |
+
]
|
85 |
+
|
86 |
+
# Process with model
|
87 |
+
text = processor.apply_chat_template(messages, tokenize=False)
|
88 |
+
inputs = processor(text=text, images=image, return_tensors="pt")
|
89 |
+
inputs = {k: v.to(model.device) for k, v in inputs.items()}
|
90 |
+
|
91 |
+
# Generate
|
92 |
+
with torch.no_grad():
|
93 |
+
outputs = model.generate(
|
94 |
+
**inputs,
|
95 |
+
max_new_tokens=256,
|
96 |
+
do_sample=False,
|
97 |
+
pad_token_id=processor.tokenizer.pad_token_id
|
98 |
+
)
|
99 |
+
|
100 |
+
# Decode output
|
101 |
+
transcription = processor.decode(outputs[0], skip_special_tokens=True)
|
102 |
+
return transcription.strip()
|
103 |
+
except Exception as e:
|
104 |
+
return f"เกิดข้อผิดพลาด: {str(e)}"
|
105 |
|
106 |
# Initialize application
|
107 |
print("กำลังเริ่มต้นแอปพลิเคชัน...")
|
108 |
if load_model_and_processor():
|
109 |
+
# Create Gradio interface
|
110 |
+
demo = gr.Interface(
|
111 |
+
fn=process_handwriting,
|
112 |
+
inputs=gr.Image(type="pil", label="อัพโหลดรูปลายมือเขียนภาษาไทย"),
|
113 |
+
outputs=gr.Textbox(label="ข้อความที่แปลงได้"),
|
114 |
+
title="Thai Handwriting Recognition",
|
115 |
+
description="อัพโหลดรูปภาพลายมือเขียนภาษาไทยเพื่อแปลงเป็นข้อความ",
|
116 |
+
examples=[["example1.jpg"], ["example2.jpg"]]
|
117 |
+
)
|
118 |
+
|
119 |
+
if __name__ == "__main__":
|
120 |
+
demo.launch()
|
121 |
else:
|
122 |
+
print("ไม่สามารถเริ่มต้นแอปพลิเคชันได้")
|