machine-learning-education

Animated educational video generator for Targeted Maximum Likelihood Estimation (TMLE) causal inference using Manim animations and Chinese TTS narration

interactive, sub-16ms vector explainer built for the DataForge 2026 Pathway Track.It proves why classical State Space Models suffer exponential associative forgetting ($\rho(A)^L$) and visualizes how Pathway's Dragon Hatchling (BDH) synaptic plasticity and BDH CQ continuous latent reasoning eliminate interference without KV-cache explosion.

A bilingual Jupyter Book that teaches reinforcement learning through failure cases and ablation experiments.

Training repo for Toy GPT: Context-2 model with attention + small domain corpus (030_analytics.txt). Attention requires scale.

Training repo for Toy GPT: Context-3 model + small neutral corpus (000_cat_dog.txt)

Training repo for Toy GPT: Context-2 model with embeddings + small domain corpus (030_analytics.txt). Much more efficient use of space.

Training repo for Toy GPT: unigram model + small structured corpus (001_animals.txt)

Training repo for Toy GPT: context-3 model + small structured corpus (001_animals.txt)

Training repo for Toy GPT: bigram + small neutral corpus (000_cat_dog.txt)

Training repo for Toy GPT: Context-2 model + small neutral corpus (000_cat_dog.txt)

Training repo for Toy GPT: context-2 model + small domain corpus (010_llm_glossary.txt)

Training repo for Toy GPT: unigram + small neutral corpus (000_cat_dog.txt)

Training repo for Toy GPT: Context-3 model with attention + small domain corpus (030_analytics.txt). Attention requires scale.

Training repo for Toy GPT: context-2 model + small structured corpus (001_animals.txt)

Training repo for Toy GPT: bigram model + small structured corpus (001_animals.txt)

Training repo for Toy GPT: bigram model + small domain corpus (010_llm_glossary.txt)

在 3D 空间拖拽积木搭建神经网络，实时形状校验 + PyTorch 真实训练 | Build neural networks with 3D blocks

An interactive 98K-parameter language model built from scratch — visualize data, prediction, loss, gradients, weight updates, pre-training, and fine-tuning.

An ~8M parameter LLM that talks like a tiny robot vacuum.

🚀 Train a custom unigram model using simple and efficient methods, enabling easy adoption for natural language processing tasks.