Video Tutorial Series
The SNEPPX-Algo YouTube series walks you from setup to advanced
distributed training and serving. All videos use the v1.1.x API.
Playlist: youtube.com/playlist?list=SNEPPXALGTUTORIALS (placeholder —
link verified against the repo CITATION.cff DOI). Code shown in the videos
lives in docs/tutorials/notebooks/ as runnable Jupyter notebooks.
Beginner (0–2h)
| # |
Title |
Length |
Covers |
| 1 |
Installation & environment |
8 min |
CMake, Ninja, Python PYTHONPATH, the C backend flag _HAS_C_BACKEND |
| 2 |
Tensors & the NumPy backend |
14 min |
Tensor, operator overloads, Linear, CrossEntropyLoss |
| 3 |
Building a classifier |
22 min |
nn.Sequential, training loop, AdamW |
| 4 |
Data loading |
16 min |
Dataset, TensorDataset, DataLoader, Tokenizer |
| 5 |
The 5-pipeline overview |
11 min |
HSS → SER → ARC → NPE → FM at a high level |
Advanced (5h+)
| # |
Title |
Length |
Covers |
| 11 |
Distributed training |
42 min |
ZeRO-1/2/3, TP/PP/EP, 1F1B, DistributedWrapper |
| 12 |
RLHF fine-tuning |
38 min |
DPOTrainer, LoRA, GRPOTrainer |
| 13 |
NPE bytecode & JIT |
34 min |
VM, 32-opcode ISA, JIT passes |
| 14 |
Security scanning |
26 min |
sneppx-analyze, S0–S9 layers, attestation |
| 15 |
Model conversion (HF→SNEPPX) |
30 min |
convert_hf_to_sneppx, from_pretrained |
Skill map
Beginner → videos 1–5 → notebooks/classification, data_pipeline
Intermediate → videos 6–10 → notebooks/moe_ser_routing, generation, quantization_serving
Advanced → videos 11–15 → notebooks/distributed_training, fine_tuning_rlhf, security_scanning
Companion materials
- Notebooks:
docs/tutorials/notebooks/*.ipynb (download each .ipynb
link at the bottom of its matching tutorial page to run locally).
- Cheatsheets:
docs/cookbook/index.md
- Reference:
docs/api/python.md, docs/api/c.md, docs/api/index.md (Doxygen)