File size: 11,196 Bytes
a7640b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
605e76c
 
 
 
 
a7640b5
 
 
 
 
 
605e76c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a7640b5
605e76c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a7640b5
 
 
 
 
 
 
605e76c
 
 
 
 
 
 
 
 
 
a7640b5
 
 
 
 
 
 
 
 
 
 
 
605e76c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a7640b5
 
605e76c
 
 
 
 
 
 
 
 
a7640b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
605e76c
a7640b5
605e76c
 
 
 
a7640b5
605e76c
 
 
a7640b5
605e76c
 
 
 
 
 
 
 
 
 
a7640b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fa58b2c
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
from flask import Flask, render_template, request, jsonify, stream_template
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch
import gc
import threading
import time
import os
from datetime import datetime
import json
import logging

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

app = Flask(__name__)
app.secret_key = os.urandom(24)

class CodeLlamaService:
    def __init__(self):
        self.model = None
        self.tokenizer = None
        self.pipeline = None
        self.is_loading = False
        self.is_loaded = False
        self.load_lock = threading.Lock()
        
    def load_model(self):
        """Load Code Llama model with memory optimization for HF Spaces"""
        if self.is_loaded or self.is_loading:
            return
            
        with self.load_lock:
            if self.is_loaded or self.is_loading:
                return
                
            self.is_loading = True
            logger.info("Loading Code Llama model...")
            
            try:
                # Use the smallest Code Llama model that fits in 16GB
                model_name = "codellama/CodeLlama-7b-Instruct-hf"
                
                # Check if CUDA is available
                device = "cuda" if torch.cuda.is_available() else "cpu"
                logger.info(f"Using device: {device}")
                
                # Load tokenizer
                self.tokenizer = AutoTokenizer.from_pretrained(
                    model_name,
                    use_fast=True,
                    trust_remote_code=True
                )
                
                # Configure model loading based on device
                if device == "cuda":
                    # GPU: Use float16 for memory efficiency
                    self.model = AutoModelForCausalLM.from_pretrained(
                        model_name,
                        torch_dtype=torch.float16,
                        low_cpu_mem_usage=True,
                        trust_remote_code=True,
                        device_map="auto"
                    )
                    torch_dtype = torch.float16
                else:
                    # CPU: Use float32 to avoid Half precision errors
                    self.model = AutoModelForCausalLM.from_pretrained(
                        model_name,
                        torch_dtype=torch.float32,
                        low_cpu_mem_usage=True,
                        trust_remote_code=True
                    )
                    # Move model to CPU explicitly
                    self.model = self.model.to('cpu')
                    torch_dtype = torch.float32
                
                # Create pipeline with appropriate settings
                if device == "cuda":
                    self.pipeline = pipeline(
                        "text-generation",
                        model=self.model,
                        tokenizer=self.tokenizer,
                        torch_dtype=torch_dtype,
                        device=0  # GPU device
                    )
                else:
                    self.pipeline = pipeline(
                        "text-generation",
                        model=self.model,
                        tokenizer=self.tokenizer,
                        device=-1  # CPU device
                    )
                
                self.is_loaded = True
                logger.info("Model loaded successfully!")
                
            except Exception as e:
                logger.error(f"Error loading model: {str(e)}")
                self.is_loaded = False
                # Clean up on failure
                if hasattr(self, 'model') and self.model is not None:
                    del self.model
                if hasattr(self, 'tokenizer') and self.tokenizer is not None:
                    del self.tokenizer
                if hasattr(self, 'pipeline') and self.pipeline is not None:
                    del self.pipeline
                gc.collect()
                if torch.cuda.is_available():
                    torch.cuda.empty_cache()
            finally:
                self.is_loading = False
    
    def generate_code(self, prompt, max_length=1024, temperature=0.3):
        """Generate code based on prompt"""
        if not self.is_loaded:
            return {"error": "Model not loaded", "code": "", "explanation": ""}
            
        try:
            # Format prompt for instruction following
            formatted_prompt = f"<s>[INST] {prompt} [/INST]"
            
            # Generate response with error handling
            generation_kwargs = {
                "max_new_tokens": max_length,
                "do_sample": True if temperature > 0 else False,
                "temperature": temperature if temperature > 0 else None,
                "top_p": 0.9 if temperature > 0 else None,
                "repetition_penalty": 1.1,
                "return_full_text": False,
                "pad_token_id": self.tokenizer.eos_token_id
            }
            
            # Remove None values to avoid warnings
            generation_kwargs = {k: v for k, v in generation_kwargs.items() if v is not None}
            
            outputs = self.pipeline(formatted_prompt, **generation_kwargs)
            
            # Extract generated text
            if isinstance(outputs, list) and len(outputs) > 0:
                if 'generated_text' in outputs[0]:
                    response = outputs[0]['generated_text']
                else:
                    response = str(outputs[0])
            else:
                response = str(outputs)
            
            response = response.strip()
            
            # Split response into code and explanation if possible
            code, explanation = self._parse_response(response)
            
            return {
                "success": True,
                "code": code,
                "explanation": explanation,
                "full_response": response
            }
            
        except Exception as e:
            logger.error(f"Error generating code: {str(e)}")
            return {"error": str(e), "code": "", "explanation": ""}
    
    def _parse_response(self, response):
        """Parse response to separate code and explanation"""
        # Try to find code blocks
        if "```" in response:
            parts = response.split("```")
            code_parts = []
            explanation_parts = []
            
            for i, part in enumerate(parts):
                if i % 2 == 1:  # Odd indices are code blocks
                    # Remove language identifier if present
                    lines = part.strip().split('\n')
                    if lines and any(lang in lines[0].lower() for lang in ['python', 'javascript', 'java', 'cpp', 'c++', 'html', 'css']):
                        code_parts.append('\n'.join(lines[1:]))
                    else:
                        code_parts.append(part.strip())
                else:  # Even indices are explanations
                    if part.strip():
                        explanation_parts.append(part.strip())
            
            code = '\n\n'.join(code_parts)
            explanation = '\n\n'.join(explanation_parts)
        else:
            # If no code blocks, try to separate by common patterns
            lines = response.split('\n')
            code_lines = []
            explanation_lines = []
            
            in_code_block = False
            for line in lines:
                # Simple heuristic to detect code vs explanation
                if (line.strip().startswith(('def ', 'class ', 'import ', 'from ', 'if ', 'for ', 'while ', 'function', 'var ', 'let ', 'const ')) or
                    line.startswith(('    ', '\t')) or
                    ('=' in line and not line.strip().startswith('#') and not line.strip().startswith('//'))):
                    code_lines.append(line)
                    in_code_block = True
                elif in_code_block and line.strip() == '':
                    code_lines.append(line)  # Keep empty lines in code blocks
                else:
                    if in_code_block and line.strip():
                        # Check if this line looks like code or explanation
                        if any(char in line for char in ['{', '}', ';', '()', '[]']) and not line.strip().endswith('.'):
                            code_lines.append(line)
                        else:
                            explanation_lines.append(line)
                            in_code_block = False
                    else:
                        explanation_lines.append(line)
                        in_code_block = False
            
            code = '\n'.join(code_lines)
            explanation = '\n'.join(explanation_lines)
        
        return code.strip(), explanation.strip()

# Initialize service
llama_service = CodeLlamaService()

@app.route('/')
def index():
    return render_template('index.html')

@app.route('/api/status')
def status():
    return jsonify({
        'is_loaded': llama_service.is_loaded,
        'is_loading': llama_service.is_loading
    })

@app.route('/api/load_model', methods=['POST'])
def load_model():
    if not llama_service.is_loaded and not llama_service.is_loading:
        threading.Thread(target=llama_service.load_model).start()
        return jsonify({'status': 'loading'})
    elif llama_service.is_loaded:
        return jsonify({'status': 'loaded'})
    else:
        return jsonify({'status': 'loading'})

@app.route('/api/generate', methods=['POST'])
def generate():
    data = request.json
    
    existing_code = data.get('existing_code', '').strip()
    instruction = data.get('instruction', '').strip()
    
    if not instruction:
        return jsonify({'error': 'Instruction is required'})
    
    # Build prompt
    if existing_code:
        prompt = f"""Here is the existing code:

```
{existing_code}
```

Instruction: {instruction}

Please provide the modified/complete code and explain what changes you made."""
    else:
        prompt = f"""Instruction: {instruction}

Please provide the code and explain what it does."""
    
    # Generate response
    result = llama_service.generate_code(
        prompt, 
        max_length=2048,
        temperature=0.3
    )
    
    return jsonify(result)

@app.route('/api/explain', methods=['POST'])
def explain_code():
    data = request.json
    code = data.get('code', '').strip()
    
    if not code:
        return jsonify({'error': 'Code is required'})
    
    prompt = f"""Please explain this code in detail:

```
{code}
```

Provide a clear explanation of what this code does, how it works, and any important details."""
    
    result = llama_service.generate_code(prompt, max_length=1024, temperature=0.1)
    
    return jsonify({
        'explanation': result.get('explanation', result.get('full_response', ''))
    })

if __name__ == '__main__':
    # Load model on startup
    threading.Thread(target=llama_service.load_model).start()
    app.run(host='0.0.0.0', port=7860, debug=False, use_reloader=False)