Hugging Face's logo

Spaces:
HuggingFaceM4
/
screenshot2html
Running on Zero

App Files Community
9
VictorSanh commited on
Commit
34db65e
·
1 Parent(s): 1056de2

fixes

Browse files
Files changed (1) hide show
  1. app.py +17 -128
app.py CHANGED
@@ -21,7 +21,7 @@ PROCESSOR = AutoProcessor.from_pretrained(
21
  token=API_TOKEN,
22
  )
23
  MODEL = AutoModelForCausalLM.from_pretrained(
24
- "HuggingFaceM4/img2html", #TODO
25
  token=API_TOKEN,
26
  trust_remote_code=True,
27
  torch_dtype=torch.bfloat16,
@@ -123,138 +123,27 @@ def model_inference(
123
 
124
  inputs = PROCESSOR.tokenizer(
125
  f"{BOS_TOKEN}<fake_token_around_image>{'<image>' * image_seq_len}<fake_token_around_image>",
126
- return_tensors="pt"
 
127
  )
128
  inputs["pixel_values"] = PROCESSOR.image_processor(
129
  [image],
130
  transform=custom_transform
131
  )
132
- inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
133
- generated_ids = MODEL.generate(**inputs, bad_words_ids=BAD_WORDS_IDS)
134
- generated_text = PROCESSOR.batch_decode(generated_ids, skip_special_tokens=True)[0]
135
- print(generated_text)
136
-
137
- CAR_COMPNAY = """<!DOCTYPE html>
138
- <html lang="en">
139
- <head>
140
- <meta charset="UTF-8">
141
- <meta name="viewport" content="width=device-width, initial-scale=1.0">
142
- <title>XYZ Car Company</title>
143
- <style>
144
- body {
145
- font-family: 'Arial', sans-serif;
146
- margin: 0;
147
- padding: 0;
148
- background-color: #f4f4f4;
149
- }
150
-
151
- header {
152
- background-color: #333;
153
- color: #fff;
154
- padding: 1em;
155
- text-align: center;
156
- }
157
-
158
- nav {
159
- background-color: #555;
160
- color: #fff;
161
- padding: 0.5em;
162
- text-align: center;
163
- }
164
-
165
- nav a {
166
- color: #fff;
167
- text-decoration: none;
168
- padding: 0.5em 1em;
169
- margin: 0 1em;
170
- }
171
-
172
- section {
173
- padding: 2em;
174
- }
175
-
176
- h2 {
177
- color: #333;
178
- }
179
-
180
- .car-container {
181
- display: flex;
182
- flex-wrap: wrap;
183
- justify-content: space-around;
184
- }
185
-
186
- .car-card {
187
- width: 300px;
188
- margin: 1em;
189
- border: 1px solid #ddd;
190
- border-radius: 5px;
191
- overflow: hidden;
192
- box-shadow: 0 2px 4px rgba(0, 0, 0, 0.1);
193
- }
194
-
195
- .car-image {
196
- width: 100%;
197
- height: 150px;
198
- object-fit: cover;
199
- }
200
-
201
- .car-details {
202
- padding: 1em;
203
- }
204
-
205
- footer {
206
- background-color: #333;
207
- color: #fff;
208
- text-align: center;
209
- padding: 1em;
210
- position: fixed;
211
- bottom: 0;
212
- width: 100%;
213
- }
214
- </style>
215
- </head>
216
- <body>
217
-
218
- <header>
219
- <h1>XYZ Car Company</h1>
220
- </header>
221
-
222
- <nav>
223
- <a href="#">Home</a>
224
- <a href="#">Models</a>
225
- <a href="#">About Us</a>
226
- <a href="#">Contact</a>
227
- </nav>
228
-
229
- <section>
230
- <h2>Our Cars</h2>
231
- <div class="car-container">
232
- <div class="car-card">
233
- <img src="car1.jpg" alt="Car 1" class="car-image">
234
- <div class="car-details">
235
- <h3>Model A</h3>
236
- <p>Description of Model A.</p>
237
- </div>
238
- </div>
239
-
240
- <div class="car-card">
241
- <img src="car2.jpg" alt="Car 2" class="car-image">
242
- <div class="car-details">
243
- <h3>Model B</h3>
244
- <p>Description of Model B.</p>
245
- </div>
246
- </div>
247
-
248
- <!-- Add more car cards as needed -->
249
- </div>
250
- </section>
251
-
252
- <footer>
253
- &copy; 2024 XYZ Car Company. All rights reserved.
254
- </footer>
255
-
256
- </body>
257
- </html>"""
258
  rendered_page = render_webpage(generated_text)
259
  return generated_text, rendered_page
260
 
 
21
  token=API_TOKEN,
22
  )
23
  MODEL = AutoModelForCausalLM.from_pretrained(
24
+ "HuggingFaceM4/img2html",
25
  token=API_TOKEN,
26
  trust_remote_code=True,
27
  torch_dtype=torch.bfloat16,
 
123
 
124
  inputs = PROCESSOR.tokenizer(
125
  f"{BOS_TOKEN}<fake_token_around_image>{'<image>' * image_seq_len}<fake_token_around_image>",
126
+ return_tensors="pt",
127
+ add_special_tokens=False,
128
  )
129
  inputs["pixel_values"] = PROCESSOR.image_processor(
130
  [image],
131
  transform=custom_transform
132
  )
133
+ inputs = {
134
+ k: v.to(DEVICE)
135
+ for k, v in inputs.items()
136
+ }
137
+ generated_ids = MODEL.generate(
138
+ **inputs,
139
+ bad_words_ids=BAD_WORDS_IDS,
140
+ max_length=4096
141
+ )
142
+ generated_text = PROCESSOR.batch_decode(
143
+ generated_ids,
144
+ skip_special_tokens=True
145
+ )[0]
146
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
147
  rendered_page = render_webpage(generated_text)
148
  return generated_text, rendered_page
149