Practical AI

Set up your Python environment and OpenAI client to send your first completion request, including installing the package, configuring the API key, and running a prompt in a Jupyter notebook.

In practical ai, learn to securely hide the OpenAI api key using the python dot env library and load it as an environment variable for production.

Learn how tokens power OpenAI's GPT interactions, including prompt, completion, and total tokens, how tokenization turns words into tokens, and how model pricing and maximum tokens influence prompts.

Learn how max tokens control OpenAI completion outputs. Distinguish prompt tokens from completion tokens and adjust max tokens in the completion API to influence results.

Learn how the stop sequence parameter controls and terminates OpenAI outputs at a desired point, with examples using the top ten populated countries and single or multi-line completions.

Explore generating multiple variations with the N parameter and the Echo setting in OpenAI completions to inspect choices and manage max token usage.

Explore the world of OpenAI models, from GPT-3 legacy to GPT-4, including embeddings, whisper, moderation, and image generation, with practical notes on tokens and pricing.

Create your first chat completion with OpenAI GPT-3.5 turbo or GPT-4 in a Google Colab notebook. Explore application programming interface setup, chat-based messaging with roles and content, and model pricing.

Explains OpenAI model pricing, compares GPT-4 and GPT-3.5 turbo costs per 1000 tokens for different context lengths, and advises using GPT-3.5 turbo for testing due to lower cost.

Learn how to structure prompts with system, user, and assistant roles in chat completions, leverage function calling, and train the model using examples for clearer sentiment classification and better context.

Learn how the temperature parameter controls creativity in chat completion, from 0 to 2 with a default of 1, enabling outputs to be more random or more focused.

Transform text, images, and categorical data into numerical embeddings that reveal relationships in vector space. Use embeddings for sentiment analysis, semantic search, and personalized recommendations.

Create a single text embedding with OpenAI using text embedding 002, turning input text into a 1536-dimensional vector to measure similarity and relatedness.

Learn to estimate embedding costs with tic token, calculate total tokens for text embeddings (text embedding ETA 002), and manage pricing for large documents.

Convert 30 to 50 company profiles into OpenAI embeddings using a pandas workflow, estimate costs, and cache results to efficiently embed business descriptions for practical AI applications.

Learn how to cache embeddings for large datasets using a local pickle file, avoiding repeated OpenAI API calls by storing and retrieving embeddings from a cache.

Discover how the open-source Lang Chain unites multiple large language models and tools to build LLM-based applications, from custom prompts to document loading and retrieval.

Set up a python environment for a simple LangChain powered application, integrate OpenAI keys, and run prompts against multiple language models to generate text.

Learn to use a prompt template from Lang chain to create reusable prompts with variables like location, and generate a final prompt to feed to a language model.

Learn how to chain outputs from large language models using lang chain and simple sequential chain, using prompts and templates to pass outputs as inputs across steps.

Set up LangChain with OpenAI in a Jupyter notebook, define an LM, and run prompts to find top pizza destinations. Use prompt templates to organize variables and reuse prompts.

Chain prompts with LangChain's chain and LMChain to pass outputs between prompts using an OpenAI model. See how the top ten food destinations example demonstrates multi-step prompts and input-output chaining.

Agents unlock internet search by connecting large language models to tools like Google Search API and Serp API through Lang Chain, enabling external information access and dynamic task execution.

Explore how to add memory to an agent by using a conversation chain to store and recall prior interactions, enabling stateful, context-aware dialogue with an LLM.

Explore the LangChain schema, focusing on text and chat messages, including system, human, and AI messages, and learn how prompt engineering shapes interactions with large language models.

Use the Langchain document object to structure unstructured data with page content and metadata, enabling AI to understand context via key-value fields like news id, category, and date.

The lecture covers three model types in long chain workflows—language models, chat models, and text embedding models—and demonstrates simple input-output with a language model and structured chat prompts.

Explore how text embedding converts text into numerical vectors that preserve semantic meaning, enabling machine understanding for NLP tasks using open embeddings.

Explore Hugging Face's open source models, datasets, and spaces, and learn to use the transformer library to interact with models, deploy via inference API, and build apps with Gradio.

Create a Hugging Face account by signing up, validating your email, and log in to access models, datasets, spaces, and generate an access token for the transformer library.

Explore how transformers enable easy access to Huggingface pre-trained models via pipelines for NLP tasks like sentiment analysis, text classification, and more, with practical Colab demonstrations.

Learn to generate text with the transformer pipeline, choose pre-trained models from Hugging Face, and control input preprocessing, prompts, and output length for multiple responses.

Learn how to use the Hugging Face pipeline for zero-shot classification with a Bart large model, classifying text against candidate labels and interpreting the resulting scores.

Define a translation pipeline using the Hugging Face transformers library, configure source English and target French, and run with the T5 base model.

Explore transformer pipeline summarization, adjust max length and minimum length, and use models like Distilbert, CNN 12 six; apply to news articles and long transcripts.

Learn how token classification assigns labels to tokens for named entity recognition, using a hugging face pipeline with bert large to identify organizations, locations, dates, and persons.

Learn to use the transformer pipeline for question answering by supplying a context and a question to produce an answer; compare extractive and abstractive QA using a practical example.

Explore image classification with transformer pipelines and vision transformer models to label images across health care, real estate, and more.

Explore using the pipeline function with pre-trained models for sentiment analysis, summarization, and translation, and learn to work with model and tokenizer APIs behind the scenes.

Learn how AutoTokenizer converts raw text into input ids for models, using pretrained tokenizers, handling tokens, attention masks, and tensors in a sentiment analysis workflow.

Explore how tokenization encodes text into tokens and input IDs using a pretrained auto tokenizer. Learn to convert tokens to IDs and decode IDs back to text with hands-on examples.

Explore the tokenizer API in transformers, converting raw text to input IDs using auto tokenizer and from_pretrained, with an emphasis on sentiment analysis pipelines and PyTorch tensors.

Learn to batch input sequences by converting sentences to tokens and input ids with auto tokenizer, handle varying lengths with padding, and use auto model for sequence classification.

Explore how to prepare text for sequence classification using auto models, auto tokenizers, and tensors: tokenize, pad, shape input IDs, manage batch dimensions, and obtain logits.

Implement a sentiment analysis model in PyTorch using the transformers pipeline, with auto tokenizer and model for sequence classification, and manage input IDs, attention masks, padding, and logits.

Create a streamlit-based Q&A app that connects to the Falcon model via Hugging Face hub and LangChain, using an API token, prompt templates, and a simple UI.

Discover microsoft pi 1.5 on Hugging Face, a fine-tuned 1.3 billion-parameter model for text and code generation, with practical Colab setup, GPU guidance, and LangChain integration.

Explore Amazon bedrock for text, image, and chat generation in a unified playground. Learn to sign in, select regions, and test base and custom models.

Explore Amazon Bedrock pricing for base and foundation models, with text and image costs per thousand tokens, and how steps influence standard versus premium images.