Skip to content
TheChatArchiveThe Chat Archive
gpt-3.5-turbo-0613· openaiHugging Face Suzhen/CodeChat

create a simple python app to inference with an h5 model using its tokenizer.pickle and…

  1. User

    create a simple python app to inference with an h5 model using its tokenizer.pickle and a text input, this is a Language model, the user should specify the sequence length and the app should output the predicted text by the model

  2. Assistant

    Sure! Here's an example of a simple Python app that uses an H5 model and its tokenizer.pickle file to generate text based on a given input sequence length:

    import tensorflow as tf
    import pickle
    
    def load_tokenizer(tokenizer_path):
        with open(tokenizer_path, 'rb') as file:
            tokenizer = pickle.load(file)
        return tokenizer
    
    def preprocess_text(text, tokenizer, seq_length):
        text_sequences = tokenizer.texts_to_sequences([text])
        text_sequences = tf.keras.preprocessing.sequence.pad_sequences(text_sequences, maxlen=seq_length,
                                                                        padding='post', truncating='post')
        return text_sequences
    
    def generate_text(model, tokenizer, seq_length, seed_text, max_words=100):
        generated_text = seed_text
        for _ in range(max_words):
            input_sequence = preprocess_text(generated_text, tokenizer, seq_length)
            predicted_word_index = model.predict_classes(input_sequence, verbose=0)
            predicted_word = tokenizer.index_word[predicted_word_index[0]]
            generated_text += ' ' + predicted_word
            if predicted_word == '.':
                break
        return generated_text
    
    # Load tokenizer
    tokenizer = load_tokenizer('tokenizer.pickle')
    
    # Load model
    model = tf.keras.models.load_model('model.h5')
    
    # Get user inputs
    seq_length = int(input("Enter the sequence length: "))
    input_text = input("Enter the input text: ")
    
    # Generate text
    generated_text = generate_text(model, tokenizer, seq_length, input_text)
    
    # Print the generated text
    print("Generated Text:")
    print(generated_text)
    

    Make sure to replace 'tokenizer.pickle' and 'model.h5' with the actual paths to your tokenizer and model files. The app prompts the user to input the desired sequence length and the input text to generate the predicted text by the language model. The generated text is printed as the output.

    Note: The generate_text function assumes that the model is a language model and uses the predict_classes function to generate the next word. If your model has a different structure or approach, you may need to modify this function accordingly.

  3. User

    I get this: Traceback (most recent call last): File "c:\Users\Dell-PC\Desktop\Projets\Model-Creator\chat-A.py", line 37, in
    generated_text = generate_text(model, tokenizer, seq_length, input_text) File "c:\Users\Dell-PC\Desktop\Projets\Model-Creator\chat-A.py", line 19, in generate_text predicted_word_index = model.predict_classes(input_sequence, verbose=0) AttributeError: 'Sequential' object has no attribute 'predict_classes'. Did you mean: 'predict_step'?

  4. Assistant

    Apologies for the confusion. The predict_classes() method has been removed in recent versions of TensorFlow. You can modify the code to use the predict() method instead and then find the index of the predicted word.

    Here's an updated version of the generate_text() function that should work correctly:

    def generate_text(model, tokenizer, seq_length, seed_text, max_words=100):
        generated_text = seed_text
        for _ in range(max_words):
            input_sequence = preprocess_text(generated_text, tokenizer, seq_length)
            predicted_word_probs = model.predict(input_sequence, verbose=0)[0]
            predicted_word_index = tf.argmax(predicted_word_probs, axis=-1).numpy()
            predicted_word = tokenizer.index_word[predicted_word_index]
            generated_text += ' ' + predicted_word
            if predicted_word == '.':
                break
        return generated_text
    

    Replace the current generate_text() function in your code with this updated version. Now it should work without any issues.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.