File size: 10,571 Bytes
5446331
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import huggingface_hub
import re
    
class LlamaManager():
    def __init__(self, llama_token = None, verbose = False):
        self.verbose = verbose
        
        if self.verbose:
            print("LlamaManager::__init__::Initializing LlamaManager")
        self.client = huggingface_hub.InferenceClient(
            "meta-llama/Meta-Llama-3.1-70B-Instruct",
            token=llama_token,
        )
        if self.verbose:
            print("LlamaManager::__init__::Initialized LlamaManager")
            
            
    def __get_items_between_tags(self, input_string, tag1, tag2):
        pattern = r'' + tag1 + '(.*?)' + tag2 + ''
        return re.findall(pattern, input_string, re.DOTALL)
        
    
    def __preprocss_for_auto_generate_questions_categories(self, available_categories):
        if self.verbose:
            print("LlamaManager::__preprocss_for_auto_generate_questions_categories::Preprocessing")
        out = ""
        for available_category in available_categories:
            out += f"[A]{available_category}[/A]"
        return out
    

    def __postprocess_for_auto_generate_questions_categories(self, out):
        if self.verbose:
            print("LlamaManager::__postprocess_for_auto_generate_questions_categories::Postprocessing")
            
        out = self.__get_items_between_tags(out, r"\[L\]", r"\[/L\]")[0]
        if not out:
            if self.verbose:
                print("LlamaManager::__postprocess_for_auto_generate_questions_categories::No content found")
            return []
        out = self.__get_items_between_tags(out, r"\[A\]", r"\[/A\]")
        if not out:
            if self.verbose:
                print("LlamaManager::__postprocess_for_auto_generate_questions_categories::No categories found")
            return []
        return out
    
        
    def auto_generate_questions_categories(
        self, 
        count = 20, 
        available_categories = ["Variables"], 
        seed = 123,
        temperature = 1.0,
        top_p = 0.9,
        frequency_penalty = 0.0
        ):
        available_content_for_assistant = self.__preprocss_for_auto_generate_questions_categories(available_categories)
        if self.verbose:
            print("LlamaManager::auto_generate_questions_categories::Generating questions categories")
        
        message_content = [
            {"role": "system", "content": "You are a synthetic data generator. You must only answer questions as a list. Each item of the list should be enclosed in [A] and [/A] tags. The list should be enclosed in [L] and [/L] tags."},
            {"role": "user", "content": f"Write me {count} basic topics for python programming"},
            {"role": "assistant", "content": f"[L]{available_content_for_assistant}"}
        ]
        
        out = self.client.chat_completion(
            messages = message_content,
            max_tokens = 1000,
            stream = False,
            seed = seed,
            temperature = temperature,
            top_p = top_p,
            frequency_penalty = frequency_penalty
        )
        
        categories = self.__postprocess_for_auto_generate_questions_categories(out.choices[0].message.content)
        if self.verbose:
            print("LlamaManager::auto_generate_questions_categories::Generated questions Categories")
        
        return categories
    
    
    def __postprocess_for_auto_generate_shots_for_category(self, out):
        if self.verbose:
            print("LlamaManager::__postprocess_for_auto_generate_shots_for_category::Postprocessing")
            
        out = self.__get_items_between_tags(out, r"\[L\]", r"\[/L\]")[0]
        if not out:
            if self.verbose:
                print("LlamaManager::__postprocess_for_auto_generate_shots_for_category::No content found")
            return []
        out = self.__get_items_between_tags(out, r"\[A\]", r"\[/A\]")
        if not out:
            if self.verbose:
                print("LlamaManager::__postprocess_for_auto_generate_shots_for_category::No questions found")
            return []
        return out

    
    def auto_generate_shots_for_category(
        self, 
        count, 
        category, 
        seed = 123,
        temperature = 1.0,
        top_p = 0.9,
        frequency_penalty = 0.0
        ):
        if self.verbose:
            print("LlamaManager::auto_generate_shots_for_category::Generating shots for category")
        
        message_content = [
            {"role": "system", "content": "You are a synthetic data generator. You must only answer questions as a list. Each item of the list should be enclosed in [A] and [/A] tags. The list should be enclosed in [L] and [/L] tags."},
            {"role": "user", "content": f"Write me 2 programming questions on the topic of For Loop in Python. The question should be of medium and hard difficulty. The question should involve use of just one function"},
            {"role": "assistant", "content": f"""[L]
             - [A]Write a program that takes a positive integer as input and computes the sum of its digits using a for loop.[/A]
             - [A]Write a program that generates a spiral matrix of size NxN, where N is always an odd number. Fill the spiral matrix with consecutive prime numbers in a clockwise spiral pattern, starting from the center of the matrix.[/A]
             """},
            {"role": "user", "content": f"Write me {count} programming questions on the topic of {category} in Python. The question should be of medium and hard difficulty. The question should involve use of just one function"},
            {"role": "assistant", "content": f"[L]"}
        ]
        
        out = self.client.chat_completion(
            messages = message_content,
            max_tokens = 1000,
            stream = False,
            seed = seed,
            temperature = temperature,
            top_p = top_p,
            frequency_penalty = frequency_penalty
        )
        
        shots = self.__postprocess_for_auto_generate_shots_for_category(out.choices[0].message.content + "[/L]")
        if self.verbose:
            print(f"LlamaManager::auto_generate_shots_for_category::Generated {count} shots for {category}")
        
        return shots
    
    
    def __preprocess_for_auto_generate_questions_from_shots(self, shots):
        if self.verbose:
            print("LlamaManager::__preprocess_for_auto_generate_questions_from_shots::Preprocessing")
        out = ""
        for shot in shots:
            out += f"[A]{shot}[/A]"
        return out

    
    def __postprocess_for_auto_generate_questions_from_shots(self, out):
        if self.verbose:
            print("LlamaManager::__postprocess_for_auto_generate_questions_from_shots::Postprocessing")
            
        out = self.__get_items_between_tags(out, r"\[L\]", r"\[/L\]")[0]
        if not out:
            if self.verbose:
                print("LlamaManager::__postprocess_for_auto_generate_questions_from_shots::No content found")
            return []
        out = self.__get_items_between_tags(out, r"\[A\]", r"\[/A\]")
        if not out:
            if self.verbose:
                print("LlamaManager::__postprocess_for_auto_generate_questions_from_shots::No questions found")
            return []
        return out
    
    
    def auto_generate_questions_from_shots(
        self,
        count,
        category,
        shots,
        seed = 123,
        temperature = 1.0,
        top_p = 0.9,
        frequency_penalty = 0.0
        ):
        available_content_for_assistant = self.__preprocess_for_auto_generate_questions_from_shots(shots)
        if self.verbose:
            print("LlamaManager::auto_generate_questions_from_shots::Generating questions from shots")
        
        message_content = [
            {"role": "system", "content": "You are a synthetic data generator. You must only answer questions as a list. Each item of the list should be enclosed in [A] and [/A] tags. The list should be enclosed in [L] and [/L] tags."},
            {"role": "user", "content": f"Write me {count} python programming questions which uses {category.lower()}"},
            {"role": "assistant", "content": f"[L]{available_content_for_assistant}"}
        ]
        
        previous_iteration_questions_count = []
        questions = []
        token_count = 1000
        while len(questions) < count:
            out = self.client.chat_completion(
                messages = message_content,
                max_tokens = token_count,
                stream = False,
                seed = seed,
                temperature = temperature,
                top_p = top_p,
                frequency_penalty = frequency_penalty
            )

            questions = self.__postprocess_for_auto_generate_questions_from_shots(out.choices[0].message.content + "[/L]")
            available_content_for_assistant = self.__preprocess_for_auto_generate_questions_from_shots(questions)
            previous_iteration_questions_count.append(len(questions))
            message_content = [
                {"role": "system", "content": "You are a synthetic data generator. You must only answer questions as a list. Each item of the list should be enclosed in [A] and [/A] tags. The list should be enclosed in [L] and [/L] tags."},
                {"role": "user", "content": f"Write me {count} python programming questions which uses {category.lower()}"},
                {"role": "assistant", "content": f"[L]{available_content_for_assistant}"}
            ]
            token_count += 500
            
            if len(previous_iteration_questions_count) > 3:
                if previous_iteration_questions_count[-1] == previous_iteration_questions_count[-2] == previous_iteration_questions_count[-3] == previous_iteration_questions_count[-4]:
                    if self.verbose:
                        print("LlamaManager::auto_generate_questions_from_shots::Generation could not be completed, stopping API calls")
                    break
        
        if self.verbose:    
            print("LlamaManager::auto_generate_questions_from_shots::Generated questions from shots")
        
        return questions

        
if __name__ == "__main__":
    llama_manager = LlamaManager("nope", True)
    categories = llama_manager.auto_generate_questions_categories(20)
    shots = llama_manager.auto_generate_shots_for_category(2, categories[3])
    questions = llama_manager.auto_generate_questions_from_shots(10, categories[3], shots, temperature = 0.5)