الدليل الكامل حول توليد البيانات الاصطناعية LLM

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load a pre-trained LLM
model_name = "gpt2-large"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Define a prompt for synthetic data generation
prompt = "Generate a customer review for a smartphone:"

# Generate synthetic data
input_ids = tokenizer.encode(prompt, return_tensors="pt")
output = model.generate(input_ids, max_length=100, num_return_sequences=1)

# Decode and print the generated text
synthetic_review = tokenizer.decode(output[0], skip_special_tokens=True)
print(synthetic_review)

prompt = """
Generate a detailed customer review for a smartphone with the following characteristics:
- Brand: {brand}
- Model: {model}
- Key features: {features}
- Rating: {rating}/5 stars

The review should be between 50-100 words and include both positive and negative aspects.

Review:
"""

brands = ["Apple", "Samsung", "Google", "OnePlus"]
models = ["iPhone 13 Pro", "Galaxy S21", "Pixel 6", "9 Pro"]
features = ["5G, OLED display, Triple camera", "120Hz refresh rate, 8K video", "AI-powered camera, 5G", "Fast charging, 120Hz display"]
ratings = [4, 3, 5, 4]

# Generate multiple reviews
for brand, model, feature, rating in zip(brands, models, features, ratings):
filled_prompt = prompt.format(brand=brand, model=model, features=feature, rating=rating)
input_ids = tokenizer.encode(filled_prompt, return_tensors="pt")
output = model.generate(input_ids, max_length=200, num_return_sequences=1)
synthetic_review = tokenizer.decode(output[0], skip_special_tokens=True)
print(f"Review for {brand} {model}:\n{synthetic_review}\n")

few_shot_prompt = """
Generate a customer support conversation between an agent (A) and a customer (C) about a product issue. Follow this format:

C: Hello, I'm having trouble with my new headphones. The right earbud isn't working.
A: I'm sorry to hear that. Can you tell me which model of headphones you have?
C: It's the SoundMax Pro 3000.
A: Thank you. Have you tried resetting the headphones by placing them in the charging case for 10 seconds?
C: Yes, I tried that, but it didn't help.
A: I see. Let's try a firmware update. Can you please go to our website and download the latest firmware?

Now generate a new conversation about a different product issue:

C: Hi, I just received my new smartwatch, but it won't turn on.
"""

# Generate the conversation
input_ids = tokenizer.encode(few_shot_prompt, return_tensors="pt")
output = model.generate(input_ids, max_length=500, num_return_sequences=1)
synthetic_conversation = tokenizer.decode(output[0], skip_special_tokens=True)
print(synthetic_conversation)

يساعد هذا النهج LLM على فهم بنية المحادثة وأسلوبها المطلوب، مما يؤدي إلى تفاعلات دعم العملاء الاصطناعية الأكثر واقعية.

from transformers import GPT2LMHeadModel, GPT2Tokenizer
import torch

model = GPT2LMHeadModel.from_pretrained("gpt2-medium")
tokenizer = GPT2Tokenizer.from_pretrained("gpt2-medium")

def generate_conditional_text(prompt, condition, max_length=100):
    input_ids = tokenizer.encode(prompt, return_tensors="pt")
    attention_mask = torch.ones(input_ids.shape, dtype=torch.long, device=input_ids.device)

    # Encode the condition
    condition_ids = tokenizer.encode(condition, add_special_tokens=False, return_tensors="pt")

    # Concatenate condition with input_ids
    input_ids = torch.cat([condition_ids, input_ids], dim=-1)
    attention_mask = torch.cat([torch.ones(condition_ids.shape, dtype=torch.long, device=condition_ids.device), attention_mask], dim=-1)

    output = model.generate(input_ids, attention_mask=attention_mask, max_length=max_length, num_return_sequences=1, no_repeat_ngram_size=2, do_sample=True, top_k=50, top_p=0.95, temperature=0.7)

    return tokenizer.decode(output[0], skip_special_tokens=True)

# Generate product descriptions with different conditions
conditions = ["Luxury", "Budget-friendly", "Eco-friendly", "High-tech"]
prompt = "Describe a backpack:"

for condition in conditions:
description = generate_conditional_text(prompt, condition)
print(f"{condition} backpack description:\n{description}\n")

import pandas as pd
from sklearn.model_selection import train_test_split
from transformers import pipeline

# Load a small real-world dataset
real_data = pd.read_csv("small_product_reviews.csv")

# Split the data
train_data, test_data = train_test_split(real_data, test_size=0.2, random_state=42)

# Initialize the text generation pipeline
generator = pipeline("text-generation", model="gpt2-medium")

def augment_dataset(data, num_synthetic_samples):
    synthetic_data = []
    for _, row in data.iterrows():
        prompt = f"Generate a product review similar to: {row['review']}\nNew review:"
        synthetic_review = generator(prompt, max_length=100, num_return_sequences=1)[0]['generated_text']
        synthetic_data.append({'review': synthetic_review,'sentiment': row['sentiment'] # Assuming the sentiment is preserved})
        if len(synthetic_data) >= num_synthetic_samples:
            break
    return pd.DataFrame(synthetic_data)

# Generate synthetic data
synthetic_train_data = augment_dataset(train_data, num_synthetic_samples=len(train_data))

# Combine real and synthetic data
augmented_train_data = pd.concat([train_data, synthetic_train_data], ignore_index=True)

print(f"Original training data size: {len(train_data)}")
print(f"Augmented training data size: {len(augmented_train_data)}")