Ollama & Local LLMs: Fine-Tune, Deploy, Build Python AI Apps

Download Ollama and install on Windows, Linux, or Mac. Use command prompt to pull and run models, list installed models, and prepare to integrate with Python.

Learn to use Ollama in Python by installing the library, importing Olima, and running model and message to get responses; explore streaming and the generate function for local models.

Build a simple stream knit project using streamlit components like st.title, st.text, and st.button, run with streamlit run, and display a prompt plus the model's response.

Demonstrates integrating the lama 3.1 model with streamlit, capturing a prompt, running generate, and displaying the model's response; explores code generation and iterative debugging for Python projects.

Learn to run code stored as a string using the exact function, handle syntax errors with try-except, import libraries, and capture the execution result for use in Python AI apps.

Learn to capture output of the exac function as a variable using io and system libraries, buffering stdout with string IO, and integrate Python code execution into models or editors.

Learn Python with Ollama and local LLMs, including Lama 3.1, by generating and explaining code, then iteratively fix errors and test in PyCharm using prompts.

Learn to pull and run the Lawa 8p model in Ollama to describe images, supply an image, and get a descriptive text, plus integrate the workflow into Streamlit.

Integrate Ollama and Streamlit to build an image describer page. Upload files like gpg, gpeg, or png, display the image, and have the model describe it in a chat-like interface.

Upload multiple images from various directories and let the model describe each image, saving files in the same directory and processing uploaded file paths.

Extract frames from a video at a two-second interval with OpenCV, save each frame, and describe them with a model. Use Lama 2.1 to summarize frame descriptions into video summary.

Learn to build a video describer that uploads a video, extracts frames, and generates one-sentence frame descriptions plus a general video explanation using local LLMs.

Learn to chat with PDFs using a local Ollama LLM, upload and extract text from a PDF, and query it with a Python‑based prompt system to get answers.

Explore converting a video to text with whisper, transcribing audio, and answering questions about the content using Ollama LLM. Build a lightweight web interface with Streamlit for interactive Q&A.

Learn to use the Langchain library to run local models with Lanchane and Olamma, install via pip, and initialize a local llm with prompts and responses.

Learn to create prompt templates and combine them with an llm chain to generate clear answers. Explore using llama/ollama models and handle single or multiple input variables for apps.

Learn to use four input variables in a prompt template with Ollama and local LLMs, wiring user name, topic, question, and instructions to create flexible, dynamic responses.

Learn chunking long texts with a recursive character text separator, tune chunk size and overlap, convert chunks to vectors, and summarize them with an llm to a concise final summary.

Learn to convert text into embeddings with the spaces library and measure similarity between two texts using cosine similarity.

Split the long text into sentences, compute each sentence’s similarity to a reference sentence using NLP, and rank results from most to least similar.

Explore building a long-text question-answering workflow with Ollama local LLMs by chunking Gutenberg stories, ranking chunks by similarity, and answering based on the most relevant passages.

Learn to chat with saved notes via chunking text and a local llm, enabling literature reviews with Streamlit-powered, text-based answers.

Build a diary chat app that saves daily entries in JSON, lets users pick dates, and answers questions by retrieving date-based diary text using similarity.

Learn to create and activate a Python virtual environment in PowerShell using PyCharm, install necessary libraries with pip, and run a Streamlit project within the isolated environment.

Learn to use local Ollama LLM with Crew AI to define agents and tasks, set roles and goals, and run a crew to generate a list-based output.

Explore using Crew AI and Ollama to build two agents - the writer and the editor - and generate a prompt-driven story, then refine it into a final enhanced narrative.

Generate recipes and analyze their nutrition with two agents: a chef and a nutritionist. The project demonstrates idea-based recipe creation, ingredient lists, and nutrition analysis using Lama 3.1.

Build a travel itinerary planner with two agents, a trip planner to generate itineraries and a local insights expert to recommend activities and dining for destinations like Nanjing and Paris.

Build a two-agent personal finance advisor that creates a house down payment saving plan with target savings, monthly contributions, and budgeting—then suggests conservative, moderate, and aggressive investment options.

Learn to download and run Hugging Face models locally, explore categories such as text generation, text-to-text, and text-to-image, and manage storage for GPU-limited setups.

Learn to download small text generation models up to 6 billion parameters for local use, run them with transformers and torch, and delete unused models via the hugging face hub.

Learn how temperature, top_k, and top_p shape model output in Ollama and local LLMs, controlling determinism, sampling, and creativity while balancing accuracy and hallucination.

Learn to use the Q and TTS text-to-speech model, select languages and speakers, apply tone instructions, and generate and save audio files with adjustable parameters.

Explore how to use a text-to-image model based on stable diffusion lightning, install required libraries, load weights from Hugging Face, and generate high-quality images in seconds via four steps.

Learn to generate text-to-video with prompts, using a 1.3 billion-parameter skyreals volume 2 model, chunked frames, all-lab history 17, 24 fps, and a diffusion pipeline.

Learn to generate videos from a start image using an image-to-video model, control prompts, and an optional end image to keep a character consistent while adjusting inference steps and pipelines.

Learn how to find tune and llms with unslought and the onslought library, compare local setup to Google Colab, and understand how parameters and weights and biases shape ai models.

Discover how AI uses 1.8 billion parameters, including weights and biases, to model input–output relationships. See how training and fine tuning adjust these parameters to improve predictions.

Explore how LLM parameters are stored in memory using float32, float16, and bf16, and how orbit quantization reduces storage with per-group scales for large models.

Learn how a local LLM tokenizer converts text to IDs, uses special and unknown tokens, and encodes prompts into tensors for chat-template inference on GPUs.

Map word IDs from the tokenizer into 2048-dimensional embeddings, preserving meaning through a learned embedding matrix, then incorporate position embeddings to capture word order for the model.

Decode end-to-end how llms process text from tokens to vectors, using tokenizer, apply rotational positioning, compute attention across 24 decoder layers, and generate next tokens via logits over 49,153-word vocabulary.

Explore how the model converts 2048-vector embeddings into 49,153 logits, applies softmax to derive token probabilities, and uses temperature and top-p (or top-k) sampling to pick the next token.

Learn end-to-end fine-tuning of local LLMs with JSON prompts and QLoRA adapters; manage overfitting via train/validation and loss-based stopping.

Learn how to run a trained model with a tokenizer, generate function, and prompts, handle training checks, and prepare deployment to Ollama for web use.

Merge adapters into base weights with the onslaught library, quantize, and prepare a local Ollama model; then sign up, push the merged model to Ollama, and let others use it.

Encourage learners to upload their fine-tuned Ollama models to the course website, share the model name with their account, and let others search and try them.

discover a local llm assisted coding tool to build python ai apps. download a 5,000-line project, install requirements, and run a streamlit app with local qn models.

Explore how the tool explains code with an explain button, adjust font sizes for code, results, and explanations, and use deep-type and analogies to learn Python libraries.

Explore how to quickly fix syntax errors and other code issues using a fix error button, AI-assisted corrections, and side-by-side code comparison in Python projects.

Learn how the LLM remembers chat history by persisting conversations in json files, using rolling summaries to manage context, and keep memory fresh.

Create a playable snake game in Python using a local llm and a simple prompt, iterating with code merges to fix errors. Explore chat history handling and prompt refinement.

Discover how planning and web research with local LLMs build Python AI apps like a snake game. Use roadmaps, code planning, and web search to improve accuracy and reduce errors.

Save project knowledge as snippets using an embedding-backed library, then inject top similar snippets into chat to reuse code across games like snake and pong.

Learn to use Ollama and local LLMs to manage and merge Python files in chat, and build a snake game with image prompts and version history.

Develop a local subtitle generator using whisper and Ollama to transcribe videos, translate English subtitles into multiple languages with the deep translator library, and organize outputs into language-specific folders.

Learn to generate multilingual subtitles efficiently, compare batch versus individual translations, optimize translation speed by chunking text, and invite you to share questions and feature suggestions.

Create a language practice app using ai, local llms, and speech tools to learn languages, customize target languages and levels, and track progress with saved sessions.