Guide
From zero to training in 15 minutes. Start with the Rust primer if you're new to Rust, or jump straight to the tutorials if you know the language.
Getting Started
Tutorials
Tensors
Creation, operations, memory management, device transfer.
Tutorial 2Autograd
Variables, gradient tracking, backward pass, no_grad.
Tutorial 3Modules
Linear, Conv2d, normalization, dropout, RNN cells, activations.
Tutorial 4Training
Losses, optimizers, gradient clipping, the full training loop.
Tutorial 5Graph Builder
The fluent API: chains, residuals, parallel branches, mapping.
Tutorial 6Advanced Graphs
Forward references, loops, gates, switches, recurrent state.
Tutorial 7Visualization
DOT/SVG output, reading diagrams, profiling overlay.
Tutorial 8Utilities
Checkpoints, gradient clipping, parameter freezing, initialization.
Tutorial 9Training Monitor
ETA, resource tracking, live web dashboard, HTML export.
Tutorial 10Graph Tree
Hierarchical composition, freeze/thaw subtrees, subgraph checkpoints, cross-boundary observation.
Tutorial 11Multi-GPU Training
Trainer::setup, El Che cadence, auto-balancing, DataLoader integration.
Tutorial 12DDP Builder
Thread-per-GPU, Local SGD, A/B-testable backends and apply policies.
Tutorial 13Data Loading
DataLoader, resident/streaming modes, VRAM-aware prefetch, DDP integration.
Tutorial 14HuggingFace Integration
Load BERT, RoBERTa, DistilBERT, ALBERT, XLM-R, or DeBERTa-v2; classify, NER, QA, fill-mask; fine-tune and round-trip back to the HF ecosystem.
Reference
DDP Reference
Apply policies, average backends, configuration knobs, troubleshooting AllReduce timeouts.
ReferenceThe floDl CLI
fdl commands, environment overlays, fdl.yml schema, libtorch management.
ReferencePorting from PyTorch
Module mapping, FlowBuilder patterns, training loop translation. AI-assisted via /port.
ReferencePyTorch Migration
Side-by-side PyTorch vs floDl for every concept, operation, and pattern.
ReferenceTroubleshooting
Real error messages and how to fix them. Build, runtime, training.