Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
nanoDiffusion
Learn how to build a nano‑sized image generator that trains in minutes on a laptop, using a two‑file implementation built on nanoGPT.
Nano sized image generator than can be trained in a couple of CPU minutes on your personal laptop.
nanoDiffusion trains/finetunes small-scale diffusion transformers for efficient image generation.
- NanoGPTNanoGPT is Andrej Karpathy's minimalist, PyTorch-based implementation of the GPT-2 transformer: a streamlined, ~300-line codebase designed for rapid training and educational clarity.NanoGPT, developed by Andrej Karpathy, is the definitive minimalist PyTorch implementation of the GPT-2 transformer architecture. It prioritizes "teeth over education," delivering a streamlined, efficient codebase: `model.py` and `train.py` are each approximately 300 lines of Python. This simplicity allows users to quickly train or fine-tune medium-sized GPTs; for example, it can reproduce the 124M parameter GPT-2 model on OpenWebText. The project is a core resource for researchers and practitioners seeking clarity, speed, and a highly hackable foundation for large language model experimentation.
- nanoDiffusionnanoDiffusion is the simplest PyTorch-based diffusion model implementation, engineered for rapid, accessible training on consumer hardware (e.g., Apple M chips).This is nanoDiffusion: a streamlined, minimal implementation of a diffusion model in PyTorch. It provides a fast-track entry into generative AI, supporting both DDPM and DDIM samplers. The architecture is specifically optimized for accessibility, featuring native acceleration support for Apple M-series chips. This efficiency allows developers to train a decent model on the MNIST dataset in just 10 to 30 minutes on a standard MacBook. We cut the complexity, focusing on core functionality and speed: a powerful tool for quick prototyping and educational use in the diffusion space.
- Stable DiffusionStable Diffusion is a latent text-to-image diffusion model: it generates detailed, photo-realistic imagery from a text prompt.Stable Diffusion is a generative AI model, developed by Stability AI, that transforms text descriptions into high-resolution digital images. Released in 2022, this deep learning system is an open-source latent diffusion model (LDM), a key distinction from proprietary competitors. It requires minimal hardware (e.g., a modest GPU with 2.4 GB VRAM) for local operation, democratizing high-end image synthesis. Beyond text-to-image generation, the technology supports image-to-image translations, inpainting (editing within an image), and outpainting (extending an image).
- PythonPython: The high-level, general-purpose language built for readability, powering everything from web backends to advanced machine learning models.Python is the high-level, general-purpose language prioritizing clear, readable syntax (via significant indentation), ensuring rapid development for any team . Its ecosystem is massive: use it for robust web development with frameworks like Django and Flask, or leverage its power in data science with libraries such as Pandas and NumPy . The Python Package Index (PyPI) provides thousands of community-contributed modules, offering immediate solutions for tasks from network programming to GUI creation . The language is actively maintained by the Python Software Foundation (PSF), with the stable release currently at Python 3.14.0 (as of November 2025) .
- CPUThe Central Processing Unit (CPU) is the core computational engine: it fetches, decodes, and executes program instructions, driving all system operations.The CPU acts as the primary processor, executing program instructions via a continuous fetch-decode-execute cycle. Key components include the Arithmetic-Logic Unit (ALU), which handles calculations and logic; the Control Unit, which orchestrates data flow; and registers, which provide high-speed temporary storage. Modern CPUs, such as the Intel Core i9 or AMD Ryzen 9, integrate multiple cores (e.g., 8-16 cores) and threads onto a single microchip, significantly boosting parallel processing capability. Clock speed (measured in GHz) and cache size (L1, L2, L3) are critical performance metrics: higher values mean faster instruction throughput and quicker data access for operating systems and applications.
- TransformerThe Transformer is a neural network architecture that uses a multi-head self-attention mechanism to process sequences in parallel, replacing slower recurrent (RNN) and convolutional (CNN) layers.The Transformer architecture, introduced in the landmark 2017 paper 'Attention Is All You Need' by Vaswani et al. (Google), revolutionized sequence-to-sequence modeling. It operates entirely on an attention mechanism (multi-head self-attention), eliminating the need for sequential processing via Recurrent Neural Networks (RNNs). This design allows for massive parallelization, drastically reducing training time and enabling the scale-up of models to billions of parameters. It is the foundational technology for all modern Large Language Models (LLMs), including BERT and the Generative Pre-trained Transformer (GPT) series, driving state-of-the-art performance across Natural Language Processing (NLP) and computer vision tasks.
Related projects
NanoBrowser: Building Open-Source Alternative to OpenAI "Operator"
Singapore
This talk demonstrates NanoBrowser, an open-source AI-powered browser extension for web automation, highlighting community contributions, practical use cases,…
Diffusion based Interior Customisation for VR Home Tours
Mumbai
This talk demonstrates using diffusion models with Controlnets and LoRAs to customize VR apartment interiors in various styles…
Semanticards
Zürich
Learn how an 80 MB sentence transformer model, embeddings, vector similarity, and user feedback generate semantically accurate flashcard answers…
Speedrunning NanoGPT training runs
Portland
This talk covers integrating differential attention layers and the Muon optimizer into NanoGPT to improve training speed and…
Stable Diffusion - Transform your casual photos into professional portraits.
Singapore
Learn how to use ComfyUI and an open‑source Stable Diffusion model to convert casual photos into polished portrait…
The Bitter Lesson - Compute vs Human Knowledge in AI Systems
New York City
Exploring how increasing computational power consistently outperforms handcrafted AI knowledge, with case studies and implications for future AI…