ml-systems-course

KO | EN

gitlite — search

#javascript #llm #go #golang #deep-learning #linux #docker #macos #cli #kubernetes #awesome-list #pytorch #ios #mcp

ml-systems-course

(★ 62)

minimal DL library in C: 24 NAIVE cuda/cpu ops, autodiff engine, python API (ops bindings/layers/models), tensor abstraction, strides, complex indexing (multi-dim slices like numpy), computation-graph visualizer, saving/loading params, primitive garbage collection, autogenerated tests, optimizers, weight initializers, etc. (no AI assistance used)

ml-systems-course 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

## 🎯 Roadmap

**ML systems + GPU programming exercise** -- build a small (but non-toy) DL stack end-to-end and learn by implementing the internals.

- 🚀 **Blackwell-optimized CUDA kernels (from scratch with explainers)** — *under active development*
- 🔍 **PyTorch internals explainer** — notes/diagrams on how core pieces work
- 📘 **Book** — a longer-form writeup of the design + lessons learned

> ⭐ star the repo to stay in the loop

## Already implemented

Minimal DL library in C:
- ⚙️ **Core:** 24 *NAIVE* cuda/cpu ops + autodiff/backprop engine
- 🧱 **Tensors:** tensor abstraction, strides/views, complex indexing (multi-dim slices like numpy)
- 🐍 **Python API:** bindings for ops, layers (built out of the ops), models (built out of the layers)
- 🧠 **Training bits:** optimizers, weight initializers, saving/loading params
- 🧪 **Tooling:** computation-graph visualizer, autogenerated tests
- 🧹 **Memory:** automatic cleanup of intermediate tensors

> built as an ML systems learning project (no AI assistance used)

# tiny-torch

## Conv-Net

```sh
rm generated/*
# define Conv-Net and train on GPU
# (you can change DEVICE macro to run on CPU instead)
nvcc models/conv_net.cu && ./a.out
# optionally, visualize model graph
cat generated/graph.txt | dot -Tsvg -o generated/conv_net.svg
# optionally, run the generated test code
python generated/codegen.py
```

## MLP

```sh
rm generated/*
# define MLP and train on CPU 
# (you can change DEVICE macro to run on GPU instead)
nvcc models/mlp.cu && ./a.out
# optionally, visualize model graph
cat generated/graph.txt | dot -Tsvg -o generated/mlp.svg
# optionally, run the generated test code
python generated/codegen.py 
```

## Setup commands

```sh
# setup evn to run the generated test code
conda create -n tiny_torch_tests python=3.11 && conda activate tiny_torch_tests
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
```

```sh
# download data
wget https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz -P ../data
tar -xvzf ../data/cifar-10-binary.tar.gz
```

ml-systems-course

ml-systems-course 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ml-systems-course

ml-systems-course 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README