| import tensorflow as tf | |
| from GPT import generate_output | |
| import numpy as np | |
| import gradio as gr | |
| from MinimalGPT_2 import MinimalGPT | |
| from subword.apply_bpe import BPE | |
| codec = open('./subword/dataset/codec.txt', encoding='utf-8') | |
| bpe = BPE(codec) | |
| output = MinimalGPT(data_path='./subword/dataset/codec.txt', | |
| learning_rate=0, | |
| output_length=11, | |
| epochs = 0, | |
| batch_size = 1, | |
| gpt_input=10, | |
| d_model=128, | |
| h=8, | |
| decoder_stacks=1, | |
| starting_chunk = 0, | |
| ending_chunk = 0, | |
| chunk_size = 1, | |
| vocabulary_start = 0, | |
| vocabulary_end = 0, | |
| save=False, | |
| load_tokenizer='./tokenizer.mgt', | |
| load_weights='./weights.mgw', | |
| save_tokenizer=None, | |
| save_weights=None, | |
| optimizer=None, | |
| inference_only = False, | |
| return_model_and_vectorizer = True, | |
| return_model_and_vectorizer_and_output = False, | |
| GPT_attention = True, | |
| TPU = False) | |
| model, vectorizer = output[0], output[1] | |
| def pad_or_slice_tensor(tensor): | |
| length = tf.shape(tensor)[0] | |
| def pad_tensor(): | |
| num_zeros = 10 - length | |
| zeros = tf.zeros((num_zeros,), dtype=tensor.dtype) | |
| padded_tensor = tf.concat([zeros, tensor], axis=0) | |
| return padded_tensor | |
| def slice_tensor(): | |
| sliced_tensor = tensor[-10:] | |
| return sliced_tensor | |
| padded_or_sliced_tensor = tf.cond( | |
| tf.less(length, 10), | |
| pad_tensor, | |
| slice_tensor | |
| ) | |
| return padded_or_sliced_tensor | |
| def generate_text(input_text): | |
| tokens = bpe.process_line(input_text) | |
| tokens = vectorizer(tokens)[:10] | |
| tokens = pad_or_slice_tensor(tokens) | |
| tokens = tf.reshape(tokens, (1, 10)) | |
| output_text = generate_output(model, vectorizer, input_sequence = tokens.numpy(), gpt_input = 10) | |
| return output_text.replace('@@ ', '') | |
| iface = gr.Interface( | |
| fn=generate_text, | |
| inputs="text", | |
| outputs="text", | |
| title="Text Generation", | |
| description="Generate text using Huggingface Transformers", | |
| theme="default" | |
| ) | |
| iface.launch() |