https://github.com/huggon1/ml-algorithm-implementations
Educational implementations for ML, DL, LLM blocks, ViT, and CUDA.
https://github.com/huggon1/ml-algorithm-implementations
cuda machine-learning numpy pytorch vision-transformer
Last synced: about 23 hours ago
JSON representation
Educational implementations for ML, DL, LLM blocks, ViT, and CUDA.
- Host: GitHub
- URL: https://github.com/huggon1/ml-algorithm-implementations
- Owner: huggon1
- Created: 2026-03-14T13:49:50.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2026-03-14T14:00:39.000Z (4 months ago)
- Last Synced: 2026-03-15T09:13:26.071Z (4 months ago)
- Topics: cuda, machine-learning, numpy, pytorch, vision-transformer
- Language: Python
- Size: 674 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# ml-algorithm-implementations
A curated collection of small algorithm and deep learning implementations collected from personal study projects.
## Highlights
- Classic ML implementations in NumPy notebooks
- Foundational neural network modules and sequence models in PyTorch
- Focused LLM building blocks such as RoPE and LoRA
- ViT and CUDA study code collected into one repository
## Scope
This repository focuses on readable educational code rather than polished production packages.
Included areas:
- Numpy implementations
- PyTorch fundamentals
- PyTorch sequence models and transformer examples
- Small LLM building blocks such as RoPE and LoRA
- MindSpore ViT experiments
- CUDA lessons and kernel demos
## Structure
```text
ml-algorithm-implementations/
numpy/
decision_tree/
kmeans/
pytorch/
llm_blocks/
mindspore/
vit/
cuda/
```
## Notes
- Some files are self-contained runnable demos.
- Some files are study-oriented reference implementations and may need small environment-specific adjustments before training.
- Large checkpoints, IDE files, caches, and unrelated materials are intentionally omitted.
## Suggested Environment
This repository mixes several stacks, so the exact dependencies depend on which folder you want to run:
- `numpy/`: Jupyter, NumPy, Matplotlib
- `pytorch/`: PyTorch
- `mindspore/`: MindSpore
- `cuda/`: CUDA toolkit and a compatible compiler toolchain
For the lightweight Python subset:
```bash
pip install -r requirements.txt
```
For CUDA lessons, use the local README files in `cuda/` as the primary entry points because they explain the lesson order and supporting images.
## Highlights
- `numpy/decision_tree` and `numpy/kmeans` keep notebook-style implementations.
- `pytorch/` collects neural network basics such as MLP, CNN, RNN, LSTM, Seq2Seq, and Transformer.
- `pytorch/llm_blocks` keeps a few focused LLM-related experiments.
- `mindspore/vit` preserves the ViT implementation in MindSpore.
- `cuda/` keeps the lesson-based CUDA exploration code and notes.