File size: 5,357 Bytes
d03c004
 
83947eb
bbce957
84eea9f
bf2f303
f3e43ed
391774c
 
bf2f303
bbce957
a421e2f
15e9cb3
bbce957
 
 
15e9cb3
bbce957
 
15e9cb3
bbce957
 
f3e43ed
bbce957
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f3e43ed
bbce957
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
15e9cb3
bf2f303
bbce957
 
 
83947eb
bf2f303
bbce957
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
pip install --upgrade transformers

import streamlit as st
import pandas as pd
import os
import torch
from transformers import GPT2Model, GPT2Tokenizer, AutoTokenizer, AutoModelForCausalLM, pipeline
from huggingface_hub import HfFolder
from io import StringIO

# Access the Hugging Face API token from environment variables
hf_token = os.getenv('HF_API_TOKEN')

if not hf_token:
    raise ValueError("Hugging Face API token is not set. Please set the HF_API_TOKEN environment variable.")
HfFolder.save_token(hf_token)

# Set environment variable to avoid floating-point errors
os.environ['TF_ENABLE_ONEDNN_OPTS'] = '0'

# Load the tokenizer and model
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model_gpt2 = GPT2Model.from_pretrained('gpt2')

# Create a pipeline for text generation using GPT-2
text_generator = pipeline("text-generation", model=model_gpt2, tokenizer=tokenizer)

# Lazy loading function for Llama-3 model
model_llama = None
tokenizer_llama = None

def load_llama_model():
    global model_llama, tokenizer_llama
    if model_llama is None:
        model_name = "meta-llama/Meta-Llama-3.1-8B"
        model_llama = AutoModelForCausalLM.from_pretrained(
            model_name,
            torch_dtype=torch.float16,  # Use FP16 for reduced memory
            token=hf_token
        )
        tokenizer_llama = AutoTokenizer.from_pretrained(model_name, token=hf_token)

# Define your prompt template
prompt_template = """\
You are an expert in generating synthetic data for machine learning models.
Your task is to generate a synthetic tabular dataset based on the description provided below.
Description: {description}
The dataset should include the following columns: {columns}
Please provide the data in CSV format with a minimum of 100 rows per generation.
Ensure that the data is realistic, does not contain any duplicate rows, and follows any specific conditions mentioned.
Example Description:
Generate a dataset for predicting house prices with columns: 'Size', 'Location', 'Number of Bedrooms', 'Price'
Example Output:
Size,Location,Number of Bedrooms,Price
1200,Suburban,3,250000
900,Urban,2,200000
1500,Rural,4,300000
...
Description:
{description}
Columns:
{columns}
Output: """

def preprocess_user_prompt(user_prompt):
    generated_text = text_generator(user_prompt, max_length=60, num_return_sequences=1)[0]["generated_text"]
    return generated_text

def format_prompt(description, columns):
    processed_description = preprocess_user_prompt(description)
    prompt = prompt_template.format(description=processed_description, columns=",".join(columns))
    return prompt

generation_params = {
    "top_p": 0.90,
    "temperature": 0.8,
    "max_new_tokens": 512,
    "return_full_text": False,
    "use_cache": False
}

def generate_synthetic_data(description, columns):
    try:
        # Load the Llama model only when generating data
        load_llama_model()
        
        # Prepare the input for the Llama model
        formatted_prompt = format_prompt(description, columns)
        
        # Tokenize the prompt
        inputs = tokenizer_llama(formatted_prompt, return_tensors="pt").to(model_llama.device)
        
        # Generate synthetic data
        with torch.no_grad():
            outputs = model_llama.generate(
                **inputs,
                max_length=512,
                top_p=generation_params["top_p"],
                temperature=generation_params["temperature"],
                num_return_sequences=1
            )
        
        # Decode the generated output
        generated_text = tokenizer_llama.decode(outputs[0], skip_special_tokens=True)
        
        # Return the generated synthetic data
        return generated_text
    except Exception as e:
        print(f"Error in generate_synthetic_data: {e}")
        return f"Error: {e}"

def generate_large_synthetic_data(description, columns, num_rows=1000, rows_per_generation=100):
    data_frames = []
    num_iterations = num_rows // rows_per_generation

    for _ in range(num_iterations):
        generated_data = generate_synthetic_data(description, columns)
        if "Error" in generated_data:
            return generated_data
        df_synthetic = process_generated_data(generated_data)
        data_frames.append(df_synthetic)
    
    return pd.concat(data_frames, ignore_index=True)

def process_generated_data(csv_data):
    data = StringIO(csv_data)
    df = pd.read_csv(data)
    return df

# Streamlit app interface
st.title("Synthetic Data Generator")
description = st.text_input("Description", "e.g., Generate a dataset for predicting students' grades")
columns = st.text_input("Columns (comma-separated)", "e.g., name, age, course, grade")

if st.button("Generate"):
    description = description.strip()
    columns = [col.strip() for col in columns.split(',')]
    df_synthetic = generate_large_synthetic_data(description, columns)
    
    if isinstance(df_synthetic, str) and "Error" in df_synthetic:
        st.error(df_synthetic)  # Display error message if any
    else:
        st.success("Synthetic Data Generated!")
        st.dataframe(df_synthetic)  # Display the generated DataFrame
        st.download_button(
            label="Download CSV",
            data=df_synthetic.to_csv(index=False),
            file_name="synthetic_data.csv",
            mime="text/csv"
        )