Delete nlpapp_4900.py

nlpapp_4900.py

@@ -1,216 +0,0 @@
-# -*- coding: utf-8 -*-
-"""nlpapp_4900.ipynb
-
-Automatically generated by Colaboratory.
-
-Original file is located at
-    https://colab.research.google.com/drive/1QE4TN1ipB5SwIVy2a7D6_bTPujHMBoDu
-"""
-
-# Import required modules for auto-reconnecting to the Colab runtime
-# Set up auto-reconnect
-import IPython
-from google.colab import output
-
-# Configure the JS code to automatically reconnect
-def restart_runtime():
-  output.clear()
-  print('Restarting runtime...')
-  display(IPython.display.Javascript('''
-    const outputArea = this;
-    const kernel = IPython.notebook.kernel;
-    const command = 'notebook_utils.restart_runtime()';
-    kernel.execute(command).then(() => {
-      outputArea.clear_output();
-      console.log('Runtime restarted. Running all cells...');
-      IPython.notebook.execute_all_cells();
-    });
-  '''))
-
-# Set the number of minutes of inactivity before auto-reconnecting
-
-minutes = 30 #@param {type: "slider", min: 1, max: 180, step: 1} # define a slider widget to set the time before auto-reconnecting
-
-# Register the auto-reconnect function
-import time
-def auto_reconnect():
-  while True:
-    print(f'Auto-reconnect in {minutes} minute(s)...')
-    time.sleep(minutes * 60 - 5) # Subtract 5 seconds to give the reconnect JS code time to run
-    restart_runtime()
-
-# Start the auto-reconnect loop in the background
-import threading
-auto_reconnect_thread = threading.Thread(target=auto_reconnect)
-auto_reconnect_thread.start()
-
-!pip install transformers torch gradio
-
-# Download and extract the Yelp review dataset: download and extract the dataset using wget and tar
-!wget https://s3.amazonaws.com/fast-ai-nlp/yelp_review_polarity_csv.tgz
-!tar -xvzf yelp_review_polarity_csv.tgz
-
-import pandas as pd
-
-train_df = pd.read_csv('yelp_review_polarity_csv/train.csv', header=None, names=['label', 'text'])
-test_df = pd.read_csv('yelp_review_polarity_csv/test.csv', header=None, names=['label', 'text'])
-
-import re
-
-# Define a function for removing HTML tags, URLs, and extra spaces, and converting the text to lowercase
-def preprocess_text(text):
-    # remove HTML tags
-    text = re.sub('<[^<]+?>', '', text)
-    # remove URLs
-    text = re.sub(r'http\S+', '', text)
-    # remove extra spaces
-    text = re.sub(r' +', ' ', text)
-    return text.strip().lower()
-
-# Apply the preprocessing function to the 'text' column of the training and test datasets    
-train_df['text'] = train_df['text'].apply(preprocess_text)
-test_df['text'] = test_df['text'].apply(preprocess_text)
-
-import torch
-import transformers
-import gradio as gr
-from transformers import TextDataset, DataCollatorForLanguageModeling, Trainer, TrainingArguments
-
-# Instantiate the tokenizer and the GPT-2 language model from the transformers library, and set the device to CUDA if available, otherwise to CPU
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-
-tokenizer = transformers.GPT2Tokenizer.from_pretrained('gpt2')
-model = transformers.GPT2LMHeadModel.from_pretrained('gpt2').to(device)
-
-# Fine-tune the model on Yelp review dataset
-train_dataset = TextDataset(
-    tokenizer=tokenizer,
-    file_path='yelp_review_polarity_csv/train.csv',
-    block_size=128,
-)
-test_dataset = TextDataset(
-    tokenizer=tokenizer,
-    file_path='yelp_review_polarity_csv/test.csv',
-    block_size=128,
-)
-data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)
-
-# Define the training arguments, instantiate the Trainer class from the transformers library, and train the model
-training_args = TrainingArguments(
-    output_dir='./results',
-    evaluation_strategy='steps',
-    eval_steps=1000,
-    save_steps=1000,
-    save_total_limit=5,
-    logging_steps=100,
-    logging_dir='./logs',
-    num_train_epochs=3,
-    per_device_train_batch_size=16,
-    per_device_eval_batch_size=32,
-    learning_rate=1e-4,
-    weight_decay=0.01,
-    gradient_accumulation_steps=2,
-    push_to_hub=False,
-    max_steps=10000,  # set a fixed number of training steps
-    # save model checkpoints at specified intervals
-    save_strategy="steps",
-)
-
-trainer = Trainer(
-    model=model,
-    args=training_args,
-    train_dataset=train_dataset,
-    data_collator=data_collator,
-    eval_dataset=test_dataset,
-)
-trainer.train()
-trainer.save_model('./gpt2_yelp_review')
-
-# Evaluate the model on the test dataset and print the perplexity score
-eval_results = trainer.evaluate(eval_dataset=test_dataset)
-print(f"Perplexity: {eval_results['eval_loss']}")
-
-import pandas as pd
-import gradio as gr
-import re
-import torch
-import transformers
-
-# Define a function for generating text based on a prompt using the fine-tuned GPT-2 model and the tokenizer
-def generate_text(prompt, length=100, theme=None, **kwargs):
-    model = transformers.GPT2LMHeadModel.from_pretrained('./gpt2_yelp_review').to(device)
-    tokenizer = transformers.GPT2Tokenizer.from_pretrained('gpt2')
-
-    # If a theme is specified, add it to the prompt as a prefix for a special token
-    if theme:
-        prompt = ' <{}> '.format(theme.strip()) + prompt.strip()
-
-    input_ids = tokenizer.encode(prompt, return_tensors='pt').to(device)
-    attention_mask = torch.ones(input_ids.shape, dtype=torch.long, device=device)
-    pad_token_id = tokenizer.eos_token_id
-
-    # Set the max length of the generated text based on the input parameter
-    max_length = length if length > 0 else 100
-
-    sample_outputs = model.generate(
-        input_ids,
-        attention_mask=attention_mask,
-        pad_token_id=pad_token_id,
-        do_sample=True,
-        max_length=max_length,
-        top_k=50,
-        top_p=0.95,
-        temperature=0.8,
-        num_return_sequences=1,
-        no_repeat_ngram_size=2,
-        repetition_penalty=1.5,
-    )
-    generated_text = tokenizer.decode(sample_outputs[0], skip_special_tokens=True)
-
-    # Post preprocessing of the generated text
-    
-    # Remove any leading and trailing quotation marks
-    generated_text = generated_text.strip('"')
-
-    # Remove leading and trailing whitespace
-    generated_text = generated_text.strip()
-
-    # Find the special token in the generated text and remove it
-    match = re.search(r'<([^>]+)>', generated_text)
-    if match:
-        generated_text = generated_text[:match.start()] + generated_text[match.end():]
-
-    # Remove any leading numeric characters and quotation marks
-    generated_text = re.sub(r'^\d+', '', generated_text)
-    generated_text = re.sub(r'^"', '', generated_text)
-
-    # Remove any newline characters from the generated text
-    generated_text = generated_text.replace('\n', '')
-
-    # Remove any other unwanted special characters
-    generated_text = re.sub(r'[^\w\s]+', '', generated_text)
-    
-    return generated_text.strip().capitalize()
-
-# Define a Gradio interface for the generate_text function, allowing users to input a prompt and generate text based on it
-iface = gr.Interface(
-    fn=generate_text,
-    inputs=['text', gr.inputs.Slider(minimum=10, maximum=100, default=50, label='Length of text'),
-            gr.inputs.Textbox(default='Food', label='Theme')],
-    outputs=[gr.outputs.Textbox(label='Generated Text')],
-    title='Yelp Review Generator',
-    description='Generate a Yelp review based on a prompt, length of text, and theme.',
-    examples=[
-        ['I had a great experience at this restaurant.', 50, 'Service'],
-        ['The service was terrible and the food was cold.', 50, 'Atmosphere'],
-        ['The food was delicious but the service was slow.', 50, 'Food'],
-        ['The ambiance was amazing and the staff was friendly.', 75, 'Service'],
-        ['The waitstaff was knowledgeable and attentive, but the noise level was a bit high.', 75, 'Atmosphere'],
-        ['The menu had a good variety of options, but the portion sizes were a bit small for the price.', 75, 'Food']
-    ],
-    allow_flagging="manual",
-    flagging_options=[("🙌", "positive"), ("😞", "negative")],
-)
-
-iface.launch(debug=False, share=True)
-