Generative AI and LLM with Python: Plus Real-World Projects

Learn how artificial neural networks simulate learning with input, hidden, and output layers, connections, and activation functions. Train with data using backpropagation and gradient descent to minimize the loss function.

Explore how artificial neural networks learn as models, from training and checkpoints to fine tuning already trained models for related tasks, and preview transformers with Python.

Master the fundamentals of transformers, including the encoding and decoding stages. Explore the self-attention mechanism that captures long-range word relationships without recurrence.

Explore transformer-based architectures for NLP and vision, including BERT, GPT, T5, XLNet, ViT, CLIP, DALL-E, and Switch Transformer, through pre-trained models and Python projects.

Explore Hugging Face, the NLP and ML company behind transformers and diffusers, hosting models and enabling pipelines for text generation, fill the mask, Q&A, summarization, translation, and chat.

Explore Hugging Face's Transformers docs and models page, filter models by task and language, review licenses and Python code examples with PyTorch.

Discover practical text generation with Python and transformers using GPT-2 medium, exploring prompts and hyperparameters like max length, temperature, top_k, top_p, and repetition_penalty for content creation and chatbots.

Learn to run a first Python generative ai example with Transformers in Google Colab and understand the code structure. Discover the four functions that power text generation with GPT-2 medium.

Explore fill the mask tasks using Bert and other transformers to complete masked texts, with top_k predictions, and apply to text generation, named entity recognition, next-word prediction, and real-world projects.

Explore filling masks in Python using a transformers pipeline with bert-base-uncased, configuring the number of predictions, prompting sentences with a mask, and displaying scored results including tokens and sequences.

Build a Python question-and-answer app that uses a provided context to answer user questions, selecting models like distilbert-base, bert, roberta, or albert, and choosing English or Portuguese.

Build a Python question-and-answer app using a transformers pipeline, prompting for context and questions in English or Portuguese and displaying ranked answers with scores.

Explore text summarization with generative AI and text generation using transformers to produce concise summaries from input text. Choose English or Portuguese models and extend to additional languages and models.

Demonstrates a python text summarization workflow with transformers, covering setup, diagram, code, and execution. Build a bilingual summarizer using english and portuguese models that prompts for text and length.

Explore translation in Python using a transformer model to convert text between languages while preserving meaning and context, and learn about multilingual models on the Hugging Face models page.

Learn to build a Python translator using transformers and sentencepiece, implementing a translator class with init and translate, tokenization, and per language-pair model loading.

Build a chatbot that simulates human-like conversation using a fine-tuned Facebook BlenderBot model with open-domain knowledge, and simulate interactions with a while loop, with a front end added later.

Build a Python chatbot by simulating Facebook BlenderBot 400M distilled, using transformers and a tokenizer for conditional generation, then tokenize input, generate and decode responses, looping until quit or exit.

Explore prompt engineering for image generation and apply seven rules: clarity, specificity, context, theme or style, model limitations, direct instructions, and avoid contradictions.

Understand how a generative adversarial network, with a generator and a discriminator, trains to produce realistic images and to distinguish real from fake ones using loss and backpropagation.

Explore three GAN architectures, starting with DCGANs trained on fashion MNIST to generate fashion images, then a celebrity face generator, and a BIGGAN model trained on ImageNet.

Generate eight random fashion MNIST images with a DCGAN in Python by loading a pre-trained model from torch hub and visualizing a normalized image grid on gpu.

Explore PGAN, Nvidia's progressive GAN variant trained on a 200,000-image celebrity database that generates realistic, non-existent faces; each run yields a different pseudo celebrity, demonstrated with code.

Demonstrate PGAN in Python by generating four fake celebrity images from a pre-trained celebrity 512 model, using gpu when available and random noise to create unique outputs.

Explore BigGAN, a DeepMind generative adversarial network trained on ImageNet with 1000 classes, capable of high quality 128 by 128 images, and learn to change the class for different subjects.

Load the imagenet pretrained biggan in python via TensorFlow Hub, generate images from a one-hot class vector and a random z, then render koalas.

Explore diffusion models that transform random images into quality visuals, audio, and video using latent diffusion and the Hugging Face library, focusing on the model, scheduler, transformer, and image processor.

Understand how the diffusion pipeline bundles the model, schedule, transformer, and image processor to generate images with diffusers, including stable diffusion, control net, and ddim variants.

Explore diffusion model tasks, from unconditional image generation to text guided image generation, text guided image to image translation, inpainting, and text guided depth to image translation, using diffusers.

Discover how auto pipeline loads right library and function to run diffusion models and generate images from prompts, including text to image, image to image, and inpainting, with Python examples.

Explore text-to-image generation with diffusers auto pipelines in Python, loading a pre-trained model on a CUDA-enabled GPU, adjusting inference steps, and rendering the output image with matplotlib.

Explore image-to-image generation with auto pipelines and diffusers, using a prompt and Mona Lisa input image to produce an image via stable diffusion 1.5 with 50 steps on gpu.

Explore auto inpainting with a two-image setup: supply a base image and a mask plus a prompt, and watch the model fill the masked area accordingly.