File size: 11,090 Bytes
4ef094d
e89b401
 
 
 
 
 
 
3e10424
38f364a
d1f9e33
 
e85540a
d1f9e33
 
 
 
 
 
38f364a
5bf0183
 
 
 
 
 
 
 
 
 
 
 
 
 
d1f9e33
e85540a
 
4ef094d
e85540a
 
 
3e10424
e85540a
e89b401
e85540a
 
 
 
 
 
 
 
 
 
 
 
d1f9e33
 
 
38f364a
d1f9e33
 
 
 
38f364a
 
e85540a
 
 
 
 
 
 
d1f9e33
e85540a
d1f9e33
 
e85540a
d1f9e33
 
e85540a
3e10424
e85540a
 
 
 
 
 
d1f9e33
 
e85540a
d1f9e33
 
3e10424
 
e85540a
 
 
 
 
 
 
 
e89b401
90a2abe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e85540a
 
90a2abe
 
 
 
 
 
e85540a
d1f9e33
 
3e10424
e85540a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d1f9e33
 
 
90a2abe
 
e85540a
90a2abe
e85540a
90a2abe
d1f9e33
 
 
 
90a2abe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e85540a
 
90a2abe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e85540a
d1f9e33
 
 
e85540a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3e10424
d1f9e33
 
90a2abe
 
e85540a
90a2abe
e85540a
90a2abe
d1f9e33
 
 
 
90a2abe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
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
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
import gradio as gr
import torch
from PIL import Image
import io
import json
import time
import os
import hashlib
import base64
from huggingface_hub import login
import traceback
import sys
import requests

# Print Python and library versions for debugging
print(f"Python version: {sys.version}")
print(f"PyTorch version: {torch.__version__}")
import transformers
print(f"Transformers version: {transformers.__version__}")

# Print token information (first few characters only for security)
token = os.environ.get("HUGGINGFACE_TOKEN", "")
if token:
    print(f"Token found: {token[:5]}...")
else:
    print("No token found in environment variables!")

# Explicitly login with your token
try:
    login(token=token)
    print("Successfully logged in to Hugging Face Hub")
except Exception as e:
    print(f"Error logging in: {e}")

# Global variables
model = None
processor = None

# Initialize Llama 4 Scout model
def load_llama4_model():
    global model, processor
    
    if model is None or processor is None:
        try:
            print("Loading Llama 4 Scout model...")
            
            # Import the correct classes for Llama 4
            from transformers import AutoProcessor, Llama4ForConditionalGeneration
            
            model_id = "meta-llama/Llama-4-Scout-17B-16E-Instruct"
            
            # Load processor and model
            processor = AutoProcessor.from_pretrained(
                model_id,
                token=token
            )
            
            # Use 4-bit quantization to reduce memory usage
            from transformers import BitsAndBytesConfig
            
            quantization_config = BitsAndBytesConfig(
                load_in_4bit=True,
                bnb_4bit_compute_dtype=torch.float16,
                bnb_4bit_quant_type="nf4"
            )
            
            model = Llama4ForConditionalGeneration.from_pretrained(
                model_id,
                token=token,
                device_map="auto",
                torch_dtype=torch.bfloat16,
                quantization_config=quantization_config
            )
            
            print("Llama 4 Scout model loaded successfully!")
            
        except Exception as e:
            print(f"Error loading Llama 4 Scout model: {e}")
            print(traceback.format_exc())
            
            # Fall back to LLaVA if Llama 4 fails
            try:
                print("Falling back to LLaVA...")
                from transformers import AutoProcessor, AutoModelForVision2Seq
                
                processor = AutoProcessor.from_pretrained("llava-hf/llava-1.5-7b-hf")
                model = AutoModelForVision2Seq.from_pretrained(
                    "llava-hf/llava-1.5-7b-hf",
                    device_map="auto"
                )
                print("LLaVA model loaded as fallback")
            except Exception as fallback_error:
                print(f"Even fallback failed: {fallback_error}")
                raise
    
    return model, processor

# Function to convert PIL Image to base64
def image_to_base64(img):
    buffered = io.BytesIO()
    img.save(buffered, format="PNG")
    img_str = base64.b64encode(buffered.getvalue()).decode()
    return f"data:image/png;base64,{img_str}"

# Simple caching mechanism
cache = {}

def compute_image_hash(image):
    """Compute a hash for an image to use as cache key"""
    # Resize to small dimensions to ensure hash is based on content, not size
    image = image.resize((100, 100), Image.LANCZOS)
    
    # Convert to bytes
    img_byte_arr = io.BytesIO()
    image.save(img_byte_arr, format='PNG')
    img_byte_arr = img_byte_arr.getvalue()
    
    # Compute hash
    return hashlib.md5(img_byte_arr).hexdigest()

def verify_document(img, doc_type, verification_info):
    """Verify a document using Llama 4 Scout"""
    if img is None:
        return "Please upload an image"
    
    # Compute image hash for caching
    image_hash = compute_image_hash(img)
    cache_key = f"verify_{image_hash}_{doc_type}"
    
    # Check cache
    if cache_key in cache:
        return f"[CACHED] {cache[cache_key]}"
    
    try:
        # Load model and processor
        model, processor = load_llama4_model()
        
        # Create prompt
        prompt = f"""This is a {doc_type} document. 
Verify if it's authentic and extract the following information: {verification_info}
Provide your analysis in a structured format."""
        
        # Process with model
        start_time = time.time()
        print(f"Starting document verification at {start_time}")
        
        # Convert image to base64 URL
        img_url = image_to_base64(img)
        
        # Create messages format
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "image", "url": img_url},
                    {"type": "text", "text": prompt},
                ]
            },
        ]
        
        # Process input using the chat template
        inputs = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors="pt",
        ).to(model.device)
        
        # Generate output
        outputs = model.generate(
            **inputs,
            max_new_tokens=300,
        )
        
        # Decode output
        result = processor.batch_decode(outputs[:, inputs["input_ids"].shape[-1]:])[0]
        
        end_time = time.time()
        print(f"Completed document verification in {end_time - start_time:.2f} seconds")
        
        # Save to cache
        cache[cache_key] = result
        
        return result
    except Exception as e:
        error_details = traceback.format_exc()
        print(f"Error in verify_document: {e}")
        print(error_details)
        return f"Error processing document: {str(e)}\n\nPlease try again with a different image or try later."

def check_workplace(img, industry):
    """Check workplace compliance using Llama 4 Scout"""
    if img is None:
        return "Please upload an image"
    
    # Compute image hash for caching
    image_hash = compute_image_hash(img)
    cache_key = f"workplace_{image_hash}_{industry}"
    
    # Check cache
    if cache_key in cache:
        return f"[CACHED] {cache[cache_key]}"
    
    try:
        # Load model and processor
        model, processor = load_llama4_model()
        
        # Create prompt
        prompt = f"""This is a workplace in the {industry} industry.
Identify any safety or compliance issues visible in this image.
Focus on:
1. Safety hazards
2. Required signage
3. Proper equipment usage
4. Workspace organization
5. Compliance with regulations

Format your response as a detailed assessment with:
- Issues found (if any)
- Severity level for each issue
- Recommendations for correction"""
        
        # Process with model
        start_time = time.time()
        print(f"Starting workplace compliance check at {start_time}")
        
        # Convert image to base64 URL
        img_url = image_to_base64(img)
        
        # Create messages format
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "image", "url": img_url},
                    {"type": "text", "text": prompt},
                ]
            },
        ]
        
        # Process input using the chat template
        inputs = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors="pt",
        ).to(model.device)
        
        # Generate output
        outputs = model.generate(
            **inputs,
            max_new_tokens=300,
        )
        
        # Decode output
        result = processor.batch_decode(outputs[:, inputs["input_ids"].shape[-1]:])[0]
        
        end_time = time.time()
        print(f"Completed workplace compliance check in {end_time - start_time:.2f} seconds")
        
        # Save to cache
        cache[cache_key] = result
        
        return result
    except Exception as e:
        error_details = traceback.format_exc()
        print(f"Error in check_workplace: {e}")
        print(error_details)
        return f"Error processing workplace image: {str(e)}\n\nPlease try again with a different image or try later."

# Create Gradio interface
with gr.Blocks(title="StaffManager AI Assistant") as demo:
    gr.Markdown("# StaffManager AI Assistant")
    gr.Markdown("This Space provides AI capabilities for StaffManager using Llama 4 Scout.")
    
    with gr.Tab("Document Verification"):
        with gr.Row():
            with gr.Column():
                doc_image = gr.Image(type="pil", label="Upload Document")
                doc_type = gr.Dropdown(
                    ["identification", "tax", "employment", "policy"], 
                    label="Document Type",
                    value="identification"
                )
                verification_info = gr.Textbox(
                    label="Verification Data (JSON)",
                    value='{"name": "John Doe", "id_number": "ABC123456"}'
                )
                verify_button = gr.Button("Verify Document")
            with gr.Column():
                doc_result = gr.Textbox(label="Verification Result", lines=10)
        
        verify_button.click(
            fn=verify_document,
            inputs=[doc_image, doc_type, verification_info],
            outputs=[doc_result]
        )
    
    with gr.Tab("Workplace Compliance"):
        with gr.Row():
            with gr.Column():
                workplace_image = gr.Image(type="pil", label="Upload Workplace Image")
                industry_type = gr.Dropdown(
                    ["retail", "restaurant", "healthcare", "manufacturing"], 
                    label="Industry",
                    value="retail"
                )
                check_button = gr.Button("Check Compliance")
            with gr.Column():
                compliance_result = gr.Textbox(label="Compliance Assessment", lines=10)
        
        check_button.click(
            fn=check_workplace,
            inputs=[workplace_image, industry_type],
            outputs=[compliance_result]
        )
    
    with gr.Tab("About"):
        gr.Markdown("""
        ## About StaffManager AI Assistant
        
        This Space uses the Llama 4 Scout model to provide AI capabilities for StaffManager:
        
        - **Document Verification**: Verify and extract information from documents
        - **Workplace Compliance**: Identify safety and compliance issues in workplace images
        
        The model is loaded on demand and results are cached for better performance.
        
        ### Model Information
        
        - Model: meta-llama/Llama-4-Scout-17B-16E-Instruct
        - Type: Multimodal (image + text)
        - Size: 17B parameters
        """)

# Launch the app
demo.launch()