File size: 7,366 Bytes
75df934
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# my_app/model_manager.py
import google.generativeai as genai
import chat.arxiv_bot.arxiv_bot_utils as utils
import json

model = None

def create_model():
    with open("apikey.txt","r") as apikey:
        key = apikey.readline()
        genai.configure(api_key=key)
    for m in genai.list_models():
        if 'generateContent' in m.supported_generation_methods:
            print(m.name)
    print("He was there")
    config = genai.GenerationConfig(max_output_tokens=2048,
                                temperature=0.7)
    safety_settings = [
        {
            "category": "HARM_CATEGORY_DANGEROUS",
            "threshold": "BLOCK_NONE",
        },
        {
            "category": "HARM_CATEGORY_HARASSMENT",
            "threshold": "BLOCK_NONE",
        },
        {
            "category": "HARM_CATEGORY_HATE_SPEECH",
            "threshold": "BLOCK_NONE",
        },
        {
            "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
            "threshold": "BLOCK_NONE",
        },
        {
            "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
            "threshold": "BLOCK_NONE",
        },
    ]
    global model
    model = genai.GenerativeModel("gemini-pro",
                                generation_config=config,
                                safety_settings=safety_settings)
    return model

def get_model():
    global model
    if model is None:
        # Khởi tạo model ở đây
        model = create_model()  # Giả sử create_model là hàm tạo model của bạn
    return model

def extract_keyword_prompt(query):
    """A prompt that return a JSON block as arguments for querying database"""

    prompt = (
        """[INST] SYSTEM: You are an assistant that choose only one action below based on guest question.

   1. If the guest question is asking for a single specific document or article with explicit title, you need to respond the information in JSON format with 2 keys "title", "author" if found any above. The authors are separated with the word 'and'. 

   2. If the guest question is asking for relevant informations about a topic, you need to respond the information in JSON format with 2 keys "keywords", "description", include a list of keywords represent the main academic topic, \

     and a description about the main topic. You may paraphrase the keywords to add more. \

   3. If the guest is not asking for any informations or documents, you need to respond with a polite answer in JSON format with 1 key "answer".

   QUESTION: '{query}'

   [/INST]

   ANSWER: 

   """
    ).format(query=query)

    return prompt

def make_answer_prompt(input, contexts):
    """A prompt that return the final answer, based on the queried context"""

    prompt = (
        """[INST] You are a library assistant that help to search articles and documents based on user's question.

  From guest's question, you have found some records and documents that may help. Now you need to answer the guest with the information found.

  If no information found in the database, you may generate some other recommendation related to user's question using your own knowledge. Each article or paper must have a link to the pdf download page.

  You should answer in a conversational form politely.

  QUESTION: '{input}'

  INFORMATION: '{contexts}'

  [/INST]

  ANSWER:

  """
    ).format(input=input, contexts=contexts)

    return prompt

def response(args):
    """Create response context, based on input arguments"""
    keys = list(dict.keys(args))
    if "answer" in keys:
        return args['answer'], None  # trả lời trực tiếp
     
    if "keywords" in keys:
        # perform query
        query_texts = args["description"]
        keywords = args["keywords"]
        results = utils.db.query_relevant(keywords=keywords, query_texts=query_texts)
        # print(results)
        ids = results['metadatas'][0]
        if len(ids) == 0:
            # go crawl some
            new_records = utils.crawl_arxiv(keyword_list=keywords, max_results=10)
            print("Got new records: ",len(new_records))
            if type(new_records) == str:
                return "Error occured, information not found", new_records
            utils.db.add(new_records)
            utils.sqldb.add(new_records)
            results = utils.db.query_relevant(keywords=keywords, query_texts=query_texts)
            ids = results['metadatas'][0]
            print("Re-queried on chromadb, results: ",ids)
        paper_id = [id['paper_id'] for id in ids]
        paper_info = utils.sqldb.query_id(paper_id)
        print(paper_info)
        records = [] # get title (2), author (3), link (6)
        result_string = ""
        if paper_info:
            for i in range(len(paper_info)):
                result_string += "Title: {}, Author: {}, Link: {}".format(paper_info[i][2],paper_info[i][3],paper_info[i][6])
                records.append([paper_info[i][2],paper_info[i][3],paper_info[i][6]])
                return result_string, records
        else:
            return "Information not found", "Information not found"
        # invoke llm and return result

    if "title" in keys:
        title = args['title']
        authors = utils.authors_str_to_list(args['author'])
        paper_info = utils.sqldb.query(title = title,author = authors)
        # if query not found then go crawl brh
        # print(paper_info)

        if len(paper_info) == 0:
            new_records = utils.crawl_exact_paper(title=title,author=authors)
            print("Got new records: ",len(new_records))
            if type(new_records) == str:
                # print(new_records)
                return "Error occured, information not found", "Information not found"
            utils.db.add(new_records)
            utils.sqldb.add(new_records)
            paper_info = utils.sqldb.query(title = title,author = authors)
            print("Re-queried on chromadb, results: ",paper_info)
        # -------------------------------------
        records = [] # get title (2), author (3), link (6)
        result_string = ""
        for i in range(len(paper_info)):
            result_string += "Title: {}, Author: {}, Link: {}".format(paper_info[i][2],paper_info[i][3],paper_info[i][6])
            records.append([paper_info[i][2],paper_info[i][3],paper_info[i][6]])
        # process results:
        if len(result_string) == 0:
            return "Information not found", "Information not found"
        return result_string, records
        # invoke llm and return result
def full_chain_single_question(input_prompt):
    try:
        first_prompt = extract_keyword_prompt(input_prompt)
        temp_answer = model.generate_content(first_prompt).text

        args = json.loads(utils.trimming(temp_answer))
        contexts, results = response(args)
        if not results:
            # print(contexts)
            return "Random question, direct return", contexts
        else:
            output_prompt = make_answer_prompt(input_prompt,contexts)
            answer = model.generate_content(output_prompt).text
            return temp_answer, answer
    except Exception as e:
        # print(e)
        return temp_answer, "Error occured: " + str(e)