Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import AutoModelForCausalLM, AutoProcessor
|
3 |
+
from PIL import Image
|
4 |
+
import requests
|
5 |
+
import gradio as gr
|
6 |
+
import spaces
|
7 |
+
|
8 |
+
model_id = "yifeihu/TB-OCR-preview-0.1"
|
9 |
+
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
10 |
+
model = AutoModelForCausalLM.from_pretrained(
|
11 |
+
model_id,
|
12 |
+
device_map="cuda",
|
13 |
+
trust_remote_code=True,
|
14 |
+
torch_dtype="auto",
|
15 |
+
attn_implementation='flash_attention_2',
|
16 |
+
load_in_4bit=True
|
17 |
+
)
|
18 |
+
processor = AutoProcessor.from_pretrained(model_id,
|
19 |
+
trust_remote_code=True,
|
20 |
+
num_crops=16
|
21 |
+
)
|
22 |
+
|
23 |
+
@spaces.GPU
|
24 |
+
def phi_ocr(image):
|
25 |
+
question = "Convert the text to markdown format."
|
26 |
+
prompt_message = [{
|
27 |
+
'role': 'user',
|
28 |
+
'content': f'<|image_1|>\n{question}',
|
29 |
+
}]
|
30 |
+
prompt = processor.tokenizer.apply_chat_template(prompt_message, tokenize=False, add_generation_prompt=True)
|
31 |
+
inputs = processor(prompt, [image], return_tensors="pt").to("cuda")
|
32 |
+
generation_args = {
|
33 |
+
"max_new_tokens": 1024,
|
34 |
+
"temperature": 0.1,
|
35 |
+
"do_sample": False
|
36 |
+
}
|
37 |
+
generate_ids = model.generate(**inputs, eos_token_id=processor.tokenizer.eos_token_id, **generation_args)
|
38 |
+
generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
|
39 |
+
response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
|
40 |
+
response = response.split("<image_end>")[0]
|
41 |
+
return response
|
42 |
+
|
43 |
+
def process_image(input_image):
|
44 |
+
return phi_ocr(input_image)
|
45 |
+
|
46 |
+
iface = gr.Interface(
|
47 |
+
fn=process_image,
|
48 |
+
inputs=gr.Image(type="pil"),
|
49 |
+
outputs="text",
|
50 |
+
title="OCR with Phi-3.5-vision-instruct",
|
51 |
+
description="Upload an image to extract and convert text to markdown format."
|
52 |
+
)
|
53 |
+
|
54 |
+
iface.launch()
|