File size: 8,111 Bytes
35ffba0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import xml.etree.ElementTree as ET
import re
import inspect
from typing import get_type_hints
import json
import datetime
import torch
import sys
import pandas as pd
from openai import OpenAI
from functions.air_quality_data_retrieval import (
    get_historical_data_for_date,
    get_historical_data_in_date_range,
    get_future_data_in_date_range,
    get_future_data_for_date,
)
from typing import Any, Dict, List


def get_type_name(t: Any) -> str:
    """Get the name of the type."""
    name = str(t)
    if "list" in name or "dict" in name:
        return name
    else:
        return t.__name__


def serialize_function_to_json(func: Any) -> str:
    """Serialize a function to JSON."""
    signature = inspect.signature(func)
    type_hints = get_type_hints(func)

    function_info = {
        "name": func.__name__,
        "description": func.__doc__,
        "parameters": {
            "type": "object",
            "properties": {}
        },
        "returns": type_hints.get('return', 'void').__name__
    }

    for name, _ in signature.parameters.items():
        param_type = get_type_name(type_hints.get(name, type(None)))
        function_info["parameters"]["properties"][name] = {"type": param_type}

    return json.dumps(function_info, indent=2)


def get_function_calling_prompt(user_query):
    fn = """{"name": "function_name", "arguments": {"arg_1": "value_1", "arg_2": value_2, ...}}"""
    example = """{"name": "get_historical_data_in_date_range", "arguments": {"date_start": "2024-01-10", "date_end": "2024-01-14"}}"""

    prompt = f"""<|im_start|>system
You are a helpful assistant with access to the following functions:

{serialize_function_to_json(get_historical_data_for_date)}

{serialize_function_to_json(get_historical_data_in_date_range)}

{serialize_function_to_json(get_future_data_for_date)}

{serialize_function_to_json(get_future_data_in_date_range)}

###INSTRUCTIONS:
- You need to choose one function to use and retrieve paramenters for this function from the user input.
- If the user query contains 'will', and specifies a single day or date, use get_future_data_in_date_range function
- If the user query contains 'will', and specifies a range of days or dates, use get_future_data_in_date_range function.
- If the user query is for future data, but only includes a single day or date, use the get_future_data_in_date_range function,
- If the user query contains 'today' or 'yesterday', use get_historical_data_for_date function.
- If the user query contains 'tomorrow', use get_future_data_in_date_range function.
- If the user query is for historical data, and specifies a range of days or dates, use use get_historical_data_for_date function.
- If the user says a day of the week, assume the date of that day is when that day next arrives.
- Do not include feature_view and model parameters.
- Provide dates STRICTLY in the YYYY-MM-DD format.
- Generate an 'No Function needed' string if the user query does not require function calling.

IMPORTANT: Today is {datetime.date.today().strftime("%A")}, {datetime.date.today()}.

To use one of there functions respond STRICTLY with:
<onefunctioncall>
    <functioncall> {fn} </functioncall>
</onefunctioncall>

###EXAMPLES

EXAMPLE 1:
- User: Hi!
- AI Assiatant: No Function needed.

EXAMPLE 2:
- User: Is this Air Quality level good or bad?
- AI Assiatant: No Function needed.

EXAMPLE 3:
- User: When and what was the minimum air quality from 2024-01-10 till 2024-01-14?
- AI Assistant:
<onefunctioncall>
    <functioncall> {example} </functioncall>
</onefunctioncall>
<|im_end|>

<|im_start|>user
{user_query}
<|im_end|>

<|im_start|>assistant"""

    return prompt


def generate_hermes(user_query: str, model_llm, tokenizer) -> str:
    """Retrieves a function name and extracts function parameters based on the user query."""

    prompt = get_function_calling_prompt(user_query)

    tokens = tokenizer(prompt, return_tensors="pt").to(model_llm.device)
    input_size = tokens.input_ids.numel()
    with torch.inference_mode():
        generated_tokens = model_llm.generate(
            **tokens,
            use_cache=True,
            do_sample=True,
            temperature=0.2,
            top_p=1.0,
            top_k=0,
            max_new_tokens=512,
            eos_token_id=tokenizer.eos_token_id,
            pad_token_id=tokenizer.eos_token_id,
        )

    return tokenizer.decode(
        generated_tokens.squeeze()[input_size:],
        skip_special_tokens=True,
    )


def function_calling_with_openai(user_query: str, client) -> str:
    """
    Generates a response using OpenAI's chat API.

    Args:
        user_query (str): The user's query or prompt.
        instructions (str): Instructions or context to provide to the GPT model.

    Returns:
        str: The generated response from the assistant.
    """

    instructions = get_function_calling_prompt(user_query).split('<|im_start|>user')[0]

    completion = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role": "system", "content": instructions},
            {"role": "user", "content": user_query},
        ]
    )

    # Extract and return the assistant's reply from the response
    if completion and completion.choices:
        last_choice = completion.choices[0]
        if last_choice.message:
            return last_choice.message.content.strip()
    return ""


def extract_function_calls(completion: str) -> List[Dict[str, Any]]:
    """Extract function calls from completion."""
    completion = completion.strip()
    pattern = r"(<onefunctioncall>(.*?)</onefunctioncall>)"
    match = re.search(pattern, completion, re.DOTALL)
    if not match:
        return None

    multiplefn = match.group(1)
    root = ET.fromstring(multiplefn)
    functions = root.findall("functioncall")

    return [json.loads(fn.text) for fn in functions]


def invoke_function(function, feature_view, weather_fg, model) -> pd.DataFrame:
    """Invoke a function with given arguments."""
    # Extract function name and arguments from input_data
    function_name = function['name']
    arguments = function['arguments']

    # Using Python's getattr function to dynamically call the function by its name and passing the arguments
    function_output = getattr(sys.modules[__name__], function_name)(
        **arguments,
        feature_view=feature_view,
        weather_fg=weather_fg,
        model=model,
    )

    if type(function_output) == str:
        return function_output

    # Round the 'pm25' value to 2 decimal places
    function_output['pm25'] = function_output['pm25'].apply(round, ndigits=2)
    return function_output


def get_context_data(user_query: str, feature_view, weather_fg, model_air_quality, model_llm=None, tokenizer=None, client=None) -> str:
    """
    Retrieve context data based on user query.

    Args:
        user_query (str): The user query.
        feature_view: Feature View for data retrieval.
        model_air_quality: The air quality model.
        tokenizer: The tokenizer.

    Returns:
        str: The context data.
    """
    if client:
        # Generate a response using LLM
        completion = function_calling_with_openai(user_query, client)

    else:
        # Generate a response using LLM
        completion = generate_hermes(
            user_query,
            model_llm,
            tokenizer,
        )

    # Extract function calls from the completion
    functions = extract_function_calls(completion)

    # If function calls were found
    if functions:
        # Invoke the function with provided arguments
        data = invoke_function(functions[0], feature_view, weather_fg, model_air_quality)

        # Return formatted data as string
        if isinstance(data, pd.DataFrame):
            return f'Air Quality Measurements:\n' + '\n'.join(
                [f'Date: {row["date"]}; Air Quality: {row["pm25"]}' for _, row in data.iterrows()]
            )
        # Return message if data is not updated
        return data

    # If no function calls were found, return an empty string
    return ''