KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#typescript #ai-agents #cli #claude-code #codex #developer-tools #ai #dsh-plugin #ai-agent #mcp #deepseek-harness #windows

flodl

★ 59 Open GitHub ↗

rust recursive deep learning framework

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<h1 align="center">floDl</h1>

A Rust-native deep learning framework built on libtorch. 
Same GPU kernels as PyTorch. No Python. No GIL. No GC. Just Rust.

<a href="#if-you-know-pytorch-you-know-flodl">PyTorch Users</a> &bull;
 <a href="https://flodl.dev/thesis">Thesis</a> &bull;
 <a href="#getting-started">Getting Started</a> &bull;
 <a href="#the-graph-builder">Graph Builder</a> &bull;
 <a href="#graph-tree-hierarchical-composition">Graph Tree</a> &bull;
 <a href="#the-training-experience">Training</a> &bull;
 <a href="#multi-gpu-training">Multi-GPU</a> &bull;
 <a href="#huggingface-integration">HuggingFace</a> &bull;
 <a href="#pytorch-parity">Parity</a> &bull;
 <a href="#performance">Benchmarks</a> &bull;
 <a href="https://github.com/flodl-labs/flodl/blob/main/ROADMAP.md">Roadmap</a> &bull;
 <a href="https://github.com/flodl-labs/flodl/blob/main/docs/pytorch_migration.md">Migration Guide</a> &bull;
 <a href="https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/13-data-loading.md">Data Loading</a>

---

> **What's new** - floDl grew a **second GPU vendor**. `flodl-sys` builds
> and links against a hipified libtorch, the public API says `gpu_*` where
> it used to say `cuda_*` (33 deprecated aliases keep existing code
> compiling), and AMD cards are enumerated from the kernel's KFD topology.
> Training on AMD is **not yet validated on hardware** - compile, link,
> provisioning and detection are. Alongside it, a run no longer needs a
> roster: `fdl join` lets a box dial into a discovery window, and
> `fdl publish` makes the controller the authority for what a run *is*,
> which is what turns a pile of cloud instances into a cohort.
> Two changes do not warn at compile time - the Rust floor is now 1.91,
> and `flodl-hf` needs owner-qualified Hub repo ids - so see
> [UPGRADE.md](https://github.com/flodl-labs/flodl/blob/main/UPGRADE.md)
> before bumping, and the
> [CHANGELOG](https://github.com/flodl-labs/flodl/blob/main/CHANGELOG.md)
> for everything else in this release.

---

## If You Know PyTorch, You Know floDl

<table>
<tr><th>PyTorch</th><th>floDl</th></tr>
<tr><td>

```python
model = nn.Sequential(
    nn.Linear(2, 16),
    nn.GELU(),
    nn.LayerNorm(16),
    nn.Linear(16, 2),
)

pred = model(x)
loss = F.mse_loss(pred, target)
loss.backward()
optimizer.step()
```

</td><td>

```rust
let model = FlowBuilder::from(Linear::new(2, 16)?)
    .through(GELU)
    .through(LayerNorm::new(16)?)
    .through(Linear::new(16, 2)?)
    .build()?;

let pred = model.forward(&x)?;
let loss = mse_loss(&pred, &target)?;
loss.backward()?;
optimizer.step()?;
```

</td></tr>
</table>

Same concepts, same names, same GPU kernels underneath. The `?` operator
replaces silent failures with compile-time error handling. `Drop` replaces the
garbage collector. The [full migration guide](https://github.com/flodl-labs/flodl/blob/main/docs/pytorch_migration.md) covers
every op, module, and pattern.

> **New to Rust?** Read [Rust for PyTorch Users](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/00-rust-primer.md) - 10 patterns in 15 minutes.

## Getting Started

**With the CLI** (recommended, no Rust needed):

```bash
curl -sL https://flodl.dev/fdl -o fdl && chmod +x fdl
./fdl setup          # detect hardware, download libtorch, configure build environment
./fdl init my-proj   # scaffold a new project with training template
```

The `fdl` script auto-downloads a pre-compiled CLI binary (~750KB, pure Rust,
no libtorch dependency). It detects your GPUs, downloads the right libtorch
variant, and configures Docker or native builds. See the [full CLI
reference](https://github.com/flodl-labs/flodl/blob/main/docs/cli/01-install.md) for all commands.

**On Windows**, run the above inside WSL2 - a WSL2 distribution is ordinary
Linux, so it gets full CUDA and multi-GPU NCCL with nothing cut down. See
[Windows / WSL2](https://github.com/flodl-labs/flodl/blob/main/docs/windows-wsl.md).

**One-liner with Docker** (no Rust, no setup):

```bash
curl -sL https://flodl.dev/init.sh | sh -s my-project
cd my-project
./fdl build   # first build (~5 min, downloads libtorch)
./fdl run     # train the model
```

**Native** - [Rust](https://rustup.rs/) 1.91+ and libtorch:

```bash
./fdl libtorch download    # auto-detects CPU, CUDA or ROCm
cargo add flodl && cargo build
```

For CUDA: `cargo add flodl --features cuda` + [CUDA toolkit](https://developer.nvidia.com/cuda-downloads).

For AMD: `cargo add flodl --features rocm` + [ROCm](https://rocm.docs.amd.com/) (Linux only). `fdl probe` lists any missing toolkit packages with the command to install them.

For Apple Silicon (Mac M1/M2/M3/M4/M5): see [docs/mac-apple-silicon.md](https://github.com/flodl-labs/flodl/blob/main/docs/mac-apple-silicon.md) — flodl runs through the Docker `dev` service (Linux arm64); a libtorch swap and `CARGO_BUILD_JOBS=2` are needed.

> **Using tch-rs or PyTorch C++?** `fdl` also works as a standalone
> libtorch manager outside of flodl: download any CPU/CUDA variant,
> switch between installs, compile from source for mixed GPU
> architectures (e.g. sm_61 + sm_120 in one build), and emit a
> machine-readable diagnostics report. No flodl buy-in required.
> See [Setup commands](https://github.com/flodl-labs/flodl/blob/main/docs/cli/02-setup-commands.md)
> and the [`flodl-cli` crate](https://crates.io/crates/flodl-cli).

Both paths generate an annotated training template. Edit `src/main.rs` to
build your model:

```rust
use flodl::*;

// The builder itself is annotated under "The Graph Builder" below.
let model = FlowBuilder::from(Linear::new(2, 16)?)
    .through(GELU)
    .also(Linear::new(16, 16)?)
    .through(Linear::new(16, 2)?)
    .build()?;

let params = model.parameters();
let mut optimizer = Adam::new(&params, 0.01);
model.train();

for (input_t, target_t) in &batches {
    let input = Variable::new(input_t.clone(), true);
    let target = Variable::new(target_t.clone(), false);

let pred = model.forward(&input)?;
    let loss = mse_loss(&pred, &target)?;

optimizer.zero_grad();
    loss.backward()?;
    clip_grad_norm(&params, 1.0)?;
    optimizer.step()?;
}
```

For framework-managed training (same code on CPU, single GPU, multi-GPU),
the universal `Trainer` takes a step closure and owns the loop:

```rust
// One step: forward + loss, returns the loss Variable.
fn train_step(model: &impl Module, batch: &[Tensor]) -> Result<Variable> {
 let input = Variable::new(batch[0].clone(), false);
 let target = Variable::new(batch[1].clone(), false);
 mse_loss(&model.forward(&input)?, &target)
}

Trainer::builder(
    |dev| build_model_on(dev),
    |params| Adam::new(params, 0.01),
    train_step,
)
    .dataset(dataset)
    .batch_size(32)
    .num_epochs(num_epochs)
    .run()?
    .join()?;
```

For an explicit loop, `Ddp::wrap` gives per-rank gradient-sync control
(the bypass tier). The cooperative tier - `Trainer::builder(...).into_worker()` -
hands you the loop body while the controller stays authoritative over
cadence, partition, eval-election, and checkpointing. See
[Training](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/04-training.md)
for all three tiers (bypass / cooperative / managed).

## The Graph Builder

floDl's fluent graph builder lets you describe complex architectures as
readable data flow - no boilerplate, no `nn.Module` subclassing.

```rust
let model = FlowBuilder::from(Linear::new(2, 16)?)
    .through(GELU)                        // activation
    .through(LayerNorm::new(16)?)         // normalization
    .also(Linear::new(16, 16)?)           // residual connection
    .through(Linear::new(16, 2)?)         // output projection
    .build()?;
```

`build()` returns a `Graph` that implements `Module` - you can nest it
inside other graphs. Things get interesting when architectures get complex:

```rust
let g = FlowBuilder::from(encoder).tag("encoded")
    .split(modules![head_a, head_b, head_c]).merge(MergeOp::Mean)
    .loop_body(refinement_block).for_n(3).tag("refined")
    .gate(router, modules![expert_a, expert_b]).using(&["encoded"])
    .switch(selector, modules![light_path, heavy_path]).using(&["refined"])
    .through(StateAdd).using(&["memory"]).tag("memory")
    .loop_body(decoder).while_cond(halt_condition, 10)
    .through(output_head)
    .build()?;
```

Every construct - `split/merge`, `also`, `loop_body`, `gate`, `switch`, `map`,
`tag/using` - composes cleanly. Forward references (`using` before `tag`) carry
state across calls, enabling recurrent architectures without special-casing.

| Method | What it does |
|--------|-------------|
| `from(m).through(m)` | Linear chain |
| `also(m)` | Residual: `input + m(input)` |
| `fork(m)` | Side branch: capture output as tag, stream continues |
| `split(modules![...]).merge(op)` | Parallel branches, merged by `Add` or `Mean` |
| `tag(name)` / `using(refs)` | Named references - backward or forward (across calls) |
| `loop_body(body).for_n(n)` | Fixed iteration with BPTT |
| `loop_body(body).while_cond` / `until_cond` | Conditional loops |
| `gate(router, modules![...])` | Soft routing - weighted combination |
| `switch(selector, modules![...])` | Hard routing - only selected branch |
| `map(body).each()` / `.over(tag)` / `.slices(n)` | Element-wise, tagged, or sliced iteration |
| `input(names)` | Auxiliary graph inputs for multi-input architectures |

See the **[Graph Builder Tutorial](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/05-graph-builder.md)** and
the [full showcase](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/showcase/).

## Graph Tree: Hierarchical Composition

This is where floDl goes beyond PyTorch. Graphs nest inside graphs with
**label-path addressing** - dot-separated paths that let you reach into any
subgraph from the root. Train components independently, compose them into
larger architectures, and control training phases declaratively.

```rust
// Build components independently
let scan = FlowBuilder::from(scan_net).tag("hidden")
    .label("scan").build()?;

let read = FlowBuilder::from(read_net).tag("confidence")
    .label("read").build()?;

let encoder = FlowBuilder::from(scan)
    .through(read)
    .label("encoder").build()?;

// Compose into full model
let model = FlowBuilder::from(encoder)
    .through(classifier)
    .build()?;
```

### Dotted paths reach anywhere

Every tag and subgraph is addressable through dotted paths from the root:

```rust
model.validate_path("encoder")?;                 // -> Subgraph
model.validate_path("encoder.scan.hidden")?;      // -> Tag (three levels deep)
model.validate_path("encoder.read.confidence")?;  // -> Tag
```

### Declarative training phases

Freeze and thaw entire subtrees by path - no manual parameter iteration:

```rust
// Phase 1: train only the classifier, encoder is frozen
model.freeze("encoder")?;
let fresh_params = model.parameters();  // only unfrozen params
let mut opt = Adam::new(&fresh_params, 1e-3);
// ... train ...

// Phase 2: thaw scan, keep read frozen (it's proven)
model.thaw("encoder.scan")?;
let mut opt = Adam::with_groups()
    .group(&model.parameters_at("encoder.scan")?, 1e-4)  // low LR
    .group(&model.parameters_at("classifier")?, 1e-3)
    .build();
```

### Subgraph checkpoints

Train a component standalone, save it, load it into a larger model:

```rust
// Pre-trained encoder saved earlier
encoder.save_checkpoint("encoder_v1.fdl.gz")?;

// Load into the composed model - namespace + hash validated
model.load_subgraph_checkpoint("encoder", "encoder_v1.fdl.gz")?;
model.freeze("encoder.read")?;  // lock what's proven
```

### Cross-boundary observation

Metrics flow up through the tree automatically:

```rust
model.record_at("encoder.scan.loss", scan_loss)?;
model.record_at("encoder.read.accuracy", read_acc)?;
model.record_scalar("total_loss", total)?;

model.flush(&[]);  // single call flushes the entire tree

// Trends across boundaries - drive training decisions
if model.trend_at("encoder.scan.loss")?.stalled(10, 1e-4) {
    model.thaw("encoder.read")?;  // scan stalled, unfreeze read
}

// Monitor sees all metrics with dotted names automatically
monitor.log(epoch, elapsed, &model);
// -> total_loss, encoder.scan.loss, encoder.read.accuracy
```

This is progressive model composition: each component is trained and
validated independently before becoming a building block in a larger
architecture. Checkpoints, metrics, and training phases compose just like
the graphs themselves.

See the full **[Graph Tree Tutorial](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/10-graph-tree.md)**.

## The Training Experience

### Training Monitor

Drop-in monitor with adaptive ETA, resource tracking, and a live web
dashboard - no external dependencies, no separate process.

```rust
use flodl::monitor::Monitor;

let mut monitor = Monitor::new(num_epochs);
monitor.serve(3000)?;  // optional: live dashboard at http://localhost:3000

for epoch in 0..num_epochs {
    let t = std::time::Instant::now();
    // ... training ...
    monitor.log(epoch, t.elapsed(), &model);  // sees entire graph tree
}
monitor.finish();
```

```
  epoch   1/100  loss=1.5264  [49ms  ETA 4.8s]
  epoch  10/100  loss=0.3817  [25ms  ETA 2.2s]  VRAM: 2.1/6.0 GB (82%)
  epoch  50/100  loss=0.0023  [24ms  ETA 1.2s]  VRAM: 2.1/6.0 GB (82%)
  epoch 100/100  loss=0.0012  [23ms]             VRAM: 2.1/6.0 GB (82%)
  training complete in 2.8s  | loss: 0.0012
```

<a href="https://flodl.dev/ddp-benchmark">
 <img src="https://raw.githubusercontent.com/flodl-labs/flodl/main/docs/dashboard.gif" alt="floDl live training dashboard - click for interactive version" width="800">
 </a>

<a href="https://flodl.dev/ddp-benchmark">Interactive DDP dashboard</a> - real data from a 200-epoch ResNet-20 run across 3 GPUs on 2 hosts, then the same view at every level of the cluster

The live dashboard updates via Server-Sent Events (no WebSocket, no npm),
tracks CPU/GPU/RAM/VRAM, and supports late join - open it mid-training and
all past epochs backfill instantly.

**One view, repeated at every level.** On a multi-GPU or multi-host run the
page becomes a portal: `root` rolls up every host, each host rolls up its
ranks, and a rank shows its own raw measurements. The breadcrumb *is* the
record path, every level is linkable (`#path=root/host-b/rank1`), and the
legend says what it is showing - `loss (mean)` and `throughput (sum)` at an
interior node, bare keys at a leaf. The page subscribes to the level you are
on, so watching a 300-rank run costs what watching a 3-rank one costs, while
alerts stay scoped to the whole subtree so a rank death in a branch you are
not looking at still reaches you.

The same records are addressable and persistable, all off by default:

```rust
Trainer::builder(model_factory, optim_factory, train_step)
    .reports_per_epoch(20)              // loss curve *between* epoch points
    .record_log("runs/records", 0)      // JSONL tree; a record's path IS its file path (0 = default 32 MiB/node cap)
    .save_dashboard("runs/dash.html")   // the whole portal as ONE offline file
    .run()?;
```

`GET /paths`, `/node?path=`, `/history?path=&n=` and `/stream?path=` (SSE)
expose the same plane to anything that speaks HTTP; a `/node` query costs
`O(children)`, not `O(cluster)`.

```rust
monitor.save_html("training_report.html");  // epoch-feed archive
monitor.export_csv("training.csv")?;         // for external analysis
```

### Observation and Trend Queries

Tags double as observation points. Collect metrics during training and use
trend queries to make programmatic training decisions:

```rust
for epoch in 0..num_epochs {
    for (input, target) in &batches {
        let pred = graph.forward(&input)?;
        graph.collect(&["hidden"])?;                 // from graph tag
        graph.record_scalar("loss", loss.item()?);   // external metric
    }
    graph.flush(&["hidden", "loss"]);

// Programmatic training control
    if graph.trend("loss").stalled(5, 1e-4) {
        optimizer.set_lr(optimizer.lr() * 0.5);      // decay LR
    }
    if graph.trend("loss").converged(5, 1e-5) {
        break;                                        // early stopping
    }
}
```

| Method | What it does |
|--------|-------------|
| `g.collect(tags)` / `g.flush(tags)` | Batch -> epoch metric aggregation |
| `g.record_scalar(tag, value)` | Inject external metrics (loss, accuracy) |
| `g.trend(tag).slope(n)` | OLS slope over last n epochs |
| `g.trend(tag).stalled(n, tol)` | Is \|slope\| below tolerance? |
| `g.trend(tag).improving(n)` | Is loss decreasing? |
| `g.trend(tag).converged(n, tol)` | Is variance below tolerance? |
| `g.trends(tags).all_improving(n)` | Group queries across branches |

### Visualization

```rust
let svg = g.svg(Some("model.svg"))?;              // architecture diagram
g.svg_with_profile(Some("profile.svg"))?;          // timing heatmap
g.plot_html("training.html", &["loss", "head"])?;  // interactive curves
```

See the **[Training Monitor Tutorial](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/09-monitor.md)** and
the **[Observation example](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/observation/)**.

## Multi-GPU Training

The **same `Trainer::builder` call** scales from CPU to N GPUs on one
host to N GPUs across many hosts. On a host with 2+ visible CUDA
devices, floDl auto-promotes to process-per-rank fan-out
automatically - zero training-loop changes. To scope which devices a
run may use, `fdl --gpus 0,1 <cmd>` (or `--gpus all`) works on any
command, at any position.

```rust
// Universal entry: works on CPU, 1 GPU, N GPUs single-host, N GPUs multi-host.
let handle = Trainer::builder(model_factory, optim_factory, train_step)
    .dataset(dataset)
    .batch_size(64)
    .num_epochs(50)
    .run()?;

let state: TrainedState = handle.join()?;
```

**ElChe - heterogeneous-rig cadence.** Mixed GPU generations? ElChe
auto-balances: the slowest GPU anchors the pace; faster ones process
proportionally more batches per averaging window, AllReduce overhead
stays bounded, the convergence guard vetoes anchor growth when
weight-space divergence rises. No configuration needed for the common
case.

**Five DDP modes, one line each.** `ElCheConfig::default()` is
`nccl_cadence()` (recommended NCCL default). Swap to any of the five
modes for A/B testing:

```rust
.elche(ElCheConfig::cpu_async().easgd_alpha(0.6))   // fastest on the reference rig
.elche(ElCheConfig::nccl_cadence())                 // default; slow rank anchors, fast ranks fill the window
.elche(ElCheConfig::nccl_sync())                    // per-batch AllReduce baseline
.elche(ElCheConfig::cpu_cadence())                  // CPU-mediated cadence (no NVLink/P2P needed)
.elche(ElCheConfig::cpu_async().bf16_wire(true))    // halve the CPU-plane payload at every hop
```

An **outer optimizer** (SlowMo or DiLoCo, applied to the consensus between
reduce rounds via `.outer_optimizer(...)`) rides on top of any CPU-averaging
mode and is what takes the accuracy crown in the benchmark below. See the
[DDP Reference](https://github.com/flodl-labs/flodl/blob/main/docs/ddp/01-reference.md)
for the factory shape.

**Multi-host clusters.** Add an `fdl.cluster.yml` next to your
`fdl.yml`, or build the topology programmatically with
`ClusterBuilder`. Then:

```bash
fdl probe              # readiness gate: GPU + libtorch + NCCL + shared-data audit
fdl @cluster train     # SSHes each worker, pre-builds, fans out
```

Heterogeneous-rig support extends to per-host libtorch variants
(`precompiled/cu128` on one host, `builds/sm61-sm120` on another),
NCCL version-skew handling via `fdl nccl build` (builds a matching
libnccl for `LD_PRELOAD`), and elastic membership (ranks can die
without aborting the run — survivors absorb the lost rank's work;
membership only shrinks, rejoin/scale-up is not yet implemented).

> **Invariant - no CUDA before `Trainer::run`.** User binaries must
> not touch libtorch's CUDA context in `main()` (no
> `gpu_device_count()`, no `Module::on_device(CUDA(_))`, no CUDA
> tensors). The launcher exits without training; touching CUDA there
> poisons spawned children's contexts on heterogeneous rigs. Use
> `flodl::sys::detect_gpus()` (loads no GPU runtime) for pre-run GPU
> queries.

See the **[Multi-GPU Tutorial](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/11-multi-gpu.md)**,
**[Heterogeneous & Multi-Host DDP](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/12-async-ddp.md)**,
**[Data Loading Tutorial](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/13-data-loading.md)**, and
**[DDP Reference](https://github.com/flodl-labs/flodl/blob/main/docs/ddp/01-reference.md)**.

### Validation suite - `ddp-bench`

The repo ships with [`ddp-bench/`](https://github.com/flodl-labs/flodl/tree/main/ddp-bench),
a standalone validation vehicle (deliberately outside the published
workspace) that reproduces published training setups (Logistic / MLP /
LeNet-5 / ResNet-20 eager and graph-built / Char-RNN / GPT-nano /
Conv-AE / OLMo-150M on MNIST, CIFAR-10, Shakespeare, olmo-mix) to build
scientifically valid solo baselines, then measures DDP/ElChe convergence
quality against them across all five DDP modes:

```bash
fdl ddp-bench --list                       # list models and modes
fdl ddp-bench quick                        # 1-epoch smoke test
fdl ddp-bench validate                     # full sweep vs structured baselines
fdl ddp-bench --model gpt-nano --mode nccl-cadence --epochs 50 --lr-scale 2
fdl ddp-bench --report runs/report.md      # convergence report from saved runs
```

Every run produces a high-frequency `Timeline` (CPU/GPU utilization, sync
events, anchor changes, idle gaps) saved as JSON / CSV / interactive HTML
under `runs/<model>/<mode>/`, and with `--save-dashboard` each cell also
leaves a self-contained `dashboard.html` portal beside it.

### Built-in datasets

The framework ships ready-to-use parsers for common benchmarks (all
implement `BatchDataSet`, plug straight into `DataLoader::from_batch_dataset`):

```rust
use flodl::data::datasets::{Cifar10, Mnist, Shakespeare};

let mnist = Mnist::parse(&images_gz, &labels_gz)?;
let cifar = Cifar10::parse(&[&batch1, &batch2, /* ... */])?;
let text  = Shakespeare::parse(&corpus, /*seq_len=*/ 128)?;
```

`ddp-bench` downloads and caches the underlying files on first run.

## HuggingFace Integration

The [`flodl-hf`](https://crates.io/crates/flodl-hf) sibling crate loads
real HuggingFace checkpoints directly into floDl, with no Python, no
ONNX detour, and no conversion script. Six BERT-family architectures
(BERT, RoBERTa, DistilBERT, ALBERT, XLM-RoBERTa, DeBERTa-v2) are wired
end-to-end with PyTorch-verified numerical parity across four task
heads each.

```rust
use flodl_hf::models::auto::AutoModelForSequenceClassification;

let clf = AutoModelForSequenceClassification::from_pretrained(
    "cardiffnlp/twitter-roberta-base-sentiment-latest",
)?;

let results = clf.predict(&["I love this framework"])?;
for (label, score) in &results[0] {
    println!("{} ({:.3})", label, score);
}
```

`AutoModel` reads `config.json`, dispatches to the right family, and
returns a typed handle. Swap the repo id for `google-bert/bert-base-uncased`,
`albert/albert-base-v2`, `FacebookAI/xlm-roberta-base`, `microsoft/deberta-v3-base`, or
any fine-tune on top of those, and the caller stays identical.

| Entry point | Task | Output |
|---|---|---|
| `AutoModel` | Backbone (hidden states) | `[batch, seq_len, hidden]` |
| `AutoModelForSequenceClassification` | Whole-text labels | `Vec<Vec<(label, score)>>` |
| `AutoModelForTokenClassification` | Per-token labels (NER) | `Vec<Vec<TokenPrediction>>` |
| `AutoModelForQuestionAnswering` | Extractive answer span | `Answer { text, start, end, score }` |
| `AutoModelForMaskedLM` | Fill-mask candidates | `Vec<(token, prob)>` (top-k) |

Three feature profiles cover deployment shapes: full Hub + tokenizer
(default), vision-only (Hub without the regex/unicode surface), and
offline `safetensors`-only (no network, no async runtime, no TLS).
Inside an existing flodl project, `fdl add flodl-hf --playground`
scaffolds a side crate with a runnable `AutoModel` example so you can
verify a real checkpoint loads before wiring it into your main code;
`fdl add flodl-hf --install` appends `flodl-hf` to your root
`Cargo.toml`, pinned to the exact version of the flodl you are running.

```bash
fdl add flodl-hf --playground   # try it: ./flodl-hf/ sandbox crate
fdl add flodl-hf --install      # wire it: pin to your root Cargo.toml
fdl flodl-hf classify           # runs AutoModel on a real fine-tune
```

Fine-tuning uses the same loop code on CPU, single GPU, or N GPUs: task
heads `impl Module` directly, so `Trainer::run` / `Trainer::builder(...)`
distribute the head transparently when more devices are available, and
`compute_loss(enc, labels)` mirrors HF Python's `model(..., labels=...).loss`
one-call shape. Round-trip the
trained head back out with `fdl flodl-hf export` then verify it loads
into HF Python's `AutoModelFor*` with `fdl flodl-hf verify-export`.

See the **[HuggingFace Integration Tutorial](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/14-flodl-hf.md)**
for the full feature matrix, per-family entry points, tokenizer usage,
local-disk loading, the fine-tune walkthrough, the export round-trip
recipe, and the 30-cell parity matrix.

## PyTorch Parity

floDl covers the modules, losses, and optimizers you actually use:

| Category | Count | Highlights |
|----------|------:|-----------|
| **NN Modules** | 30+ | `Linear`, `Conv1d`/`2d`/`3d` + transpose, `GRU`/`LSTM`, `MultiheadAttention`, `Bilinear`, all norms (`Layer`/`RMS`/`Group`/`Batch`/`Instance`), all pooling, `Embedding`/`EmbeddingBag`, `PixelShuffle`, `Upsample`, `Unfold`/`Fold` |
| **Activations** | 17 | `ReLU`, `LeakyReLU`, `ELU`, `GELU`, `SiLU`, `Mish`, `SELU`, `Softplus`, `Hardswish`, `PReLU`, `Softmax`, ... |
| **Losses** | 15 | MSE, CrossEntropy, BCE, NLL, CTC, Focal, Triplet, KLDiv, SmoothL1, Cosine, Hinge, Margin, Poisson, ... |
| **Optimizers** | 7 | `SGD`, `Adam`, `AdamW`, `RMSprop`, `Adagrad`, `RAdam`, `NAdam` - all with parameter groups |
| **Schedulers** | 8 | Step, Cosine, Exponential, MultiStep, OneCycle, Cyclic, Warmup (composable), Plateau |
| **Init** | 9 | Xavier, Kaiming, orthogonal, truncated normal, uniform, normal |
| **Tensor Ops** | 100+ | Full arithmetic, trig, reductions, shape, indexing, comparisons, fused ops |
| **Autograd** | 90+ | Differentiable backward for every op above |

Fused Adam/AdamW on CUDA (single kernel for all parameters). Fused gradient
clipping via foreach ops. Mixed precision with `AutocastGuard` + `GradScaler`.
CUDA Graphs for replay-based training.

The [full migration guide](https://github.com/flodl-labs/flodl/blob/main/docs/pytorch_migration.md) has side-by-side
code for every op, module, and pattern.

## Performance

Same CUDA kernels as PyTorch - the difference comes from what happens
*between* kernel launches. Ten models, ten interleaved rounds, locked GPU
clocks (RTX 5060 Ti, v0.3.0 vs PyTorch 2.10.0):

| Model | PyTorch | flodl | Delta |
|---|---:|---:|---:|
| transformer | 3183.0 ms | 2199.8 ms | **-31%** |
| mlp | 291.1 ms | 207.0 ms | **-29%** |
| residual_tower | 406.9 ms | 309.7 ms | **-24%** |
| feedback_fixed | 275.3 ms | 231.3 ms | **-16%** |
| gated_routing | 248.0 ms | 217.3 ms | **-12%** |
| iterative_refine | 230.7 ms | 206.0 ms | **-11%** |
| gru_seq | 1105.1 ms | 1057.5 ms | **-4%** |
| conv_autoenc | 398.2 ms | 395.3 ms | -1% |
| lstm_seq | 692.3 ms | 692.3 ms | 0% |
| convnet | 1298.0 ms | 1298.2 ms | 0% |

Wins 8 of 10, ties 2, zero regressions. The ties (convnet, lstm_seq) are
compute-bound - both frameworks saturate the GPU, confirming identical
CUDA kernels. The gap appears where framework overhead matters:
dispatch-bound architectures (transformer -31%, mlp -29%), graph routing
(residual_tower -24%), and recurrent loops (feedback_fixed -16%).

**[Benchmark Report](https://github.com/flodl-labs/flodl/blob/main/docs/benchmark.md)** |
[Interactive dashboard](https://flodl.dev/benchmark)

### Multi-GPU (DDP)

ResNet-20 on CIFAR-10, 200 epochs - three mismatched GPUs spanning two
hosts (an RTX 5060 Ti on one, two GTX 1060s in a VM on the other, the
slowest behind a PCIe x1 riser), coordinated over TCP as a real cluster.
Published reference: 91.25%
([He et al. 2015](https://arxiv.org/abs/1512.03385), Table 6):

| Mode | Eval | vs Published | Time | vs Solo-0 |
|---|---:|---:|---:|---:|
| solo-0 (fast GPU only) | 91.46% | +0.21% | 696s | - |
| cpu-async-diloco | **92.29%** | **+1.04%** | 500s | 1.39x |
| cpu-async | **91.56%** | **+0.31%** | 498s | 1.40x |
| cpu-cadence | **91.79%** | **+0.54%** | 503s | 1.38x |
| nccl-cadence | **91.78%** | **+0.53%** | 512s | 1.36x |

Every ElChe mode surpasses published accuracy while finishing faster
than the fast GPU alone, and the DiLoCo outer optimizer takes the
accuracy crown while stopping at twice the solo run's training loss:
each replica only sees its own data partition, so replica-private
memorization is averaged away every round while shared structure
survives. 200 epochs is where ElChe's proportional scheduling has room
to calibrate - shorter models (logistic through gpt-nano) confirm DDP
convergence across architectures.

**[DDP Benchmark Report](https://github.com/flodl-labs/flodl/blob/main/docs/ddp-benchmark.md)** -
full results for 8 models across all five DDP modes plus solo baselines,
seeded and reproducible at initialization

## Why Rust for Deep Learning?

**Deterministic memory.** Python adds ~3-5 us of framework overhead per GPU
op. Go's GC can't manage VRAM - an [earlier Go implementation](https://github.com/fab2s/goDl)
required 5 phases of lifecycle management (refcounting, GC callbacks, VRAM
budgets, pending-free queues). Rust replaces all of that with
`impl Drop for Tensor`. Memory is freed the instant a tensor leaves scope.

**Zero-cost safety.** Every op returns `Result<T>` - no silent failures.
Ownership ensures tensors are freed exactly once. The borrow checker
prevents data races at compile time.

**Same GPU kernels.** floDl binds libtorch - the C++ library under
PyTorch. CUDA, cuBLAS, cuDNN are identical. floDl replaces the dispatch
path, autograd tracking, and graph execution.

## Features Reference

<details>
<summary>Training Tools</summary>

| Tool | What it does |
|------|-------------|
| `clip_grad_norm` / `clip_grad_value` | Fused gradient clipping (2 kernels total via foreach ops) |
| `save_checkpoint` / `load_checkpoint` | Named `.fdl` checkpoints, structural hash, partial loading, `LoadReport` |
| `save_state_file` / `load_state_file` | Optimizer state save/load (`.optim`, self-identifying header, gzip-aware, atomic) |
| `migrate_checkpoint` | Remap parameter names across versions |
| `migrate_optim_state_file` | Convert an old optimizer `.optim` file to the current format |
| `Parameter::freeze` / `unfreeze` | Per-parameter gradient control |
| `GradScaler` | Dynamic loss scaling for fp16 training |
| `cast_parameters` | Cast model parameters to any dtype |
| `CpuWorker` / `ModelSnapshot` | Background checkpoint saving |
| `GpuGraph` | Capture/replay training steps for fixed-shape models |

</details>

<details>
<summary>Module Traits</summary>

Beyond `forward`/`parameters`, `Module` provides optional methods the graph
recognizes automatically:

| Method | What happens |
|--------|-------------|
| `as_named_input()` | `using()` refs arrive as a named map |
| `reset()` | Loops auto-call before iterating - clears per-forward state |
| `detach_state()` | Break gradient chains on retained state |
| `sub_modules()` | Recursive device placement, training mode, parameter collection |

</details>

<details>
<summary>Build Profiles</summary>

```toml
# Optimize floDl in dev builds - your code stays fast to compile.
[profile.dev.package.flodl]
opt-level = 3

[profile.dev.package.flodl-sys]
opt-level = 3

# Release: cross-crate optimization for maximum throughput.
[profile.release]
lto = "thin"
codegen-units = 1
```

| Profile | flodl | Your code | Typical rebuild |
|---------|-------|-----------|-----------------|
| `cargo build` | `-O3` (cached) | `-O0` (fast) | < 2s |
| `cargo build --release` | `-O3` + LTO | `-O3` + LTO | full link |

</details>

<details>
<summary>Multi-GPU (DDP)</summary>

| Component | What it does |
|-----------|-------------|
| `Trainer::builder(...).run()` | Universal entry. Same call scales from CPU to multi-host cluster. |
| `Trainer::run(..., TrainerConfig)` | Config-bag form - same launcher, data-driven setup. |
| `Trainer::builder(...).into_worker()` | Cooperative tier - you own the loop body, the controller stays authoritative over cadence, partition, eval-election and checkpointing. |
| `ElCheMode` | `NcclSync/Cadence`, `CpuSync/Cadence/Async`. Default `NcclCadence`. |
| `ElCheConfig` | Anchor tuning, partition ratios, convergence guard, EASGD, meta-controller. |
| `TrainerConfig` | Umbrella: dataset, callbacks, checkpointing, resume, cluster topology. |
| `ClusterBuilder` | Programmatic cluster construction (mirrors `fdl.cluster.yml`). |
| `flodl::sys::detect_gpus` | GPU detection that loads no GPU runtime; vendor-plural (nvidia-smi / KFD), scoped to the build's vendor. Canonical pre-`Trainer::run` query. |
| `TrendGuard` / `MsfGuard` / `NoGuard` | Convergence guards - TrendGuard is default. Guard is authoritative over `overhead_target`. |
| `EpochCallbackPolicy` | `Rank(global)` or `Fastest` (default - cost-aware, free-compute on heterogeneous rigs). |
| `NcclComms` / `NcclRankComm` / `NcclAbortHandle` | Low-level NCCL when you need it. Init-on-main + `split()` everywhere. |
| `GpuEvent` / `GpuStream` / `StreamGuard` | Async GPU-CPU pipeline, timing. |
| `Ddp::wrap(&model, device, rank, &rdv)` | Bypass tier - manual thread-based per-rank gradient sync, for GAN / RL / explicit replica control. Production multi-GPU auto-promotes to process-per-rank instead. |

</details>

### Numerical Verification

Every differentiable path is verified against finite-difference gradients:
- 117 autograd op-level checks (every op + compositions)
- Module-level checks (every NN module, input + parameter gradients)
- Exact optimizer step verifications (SGD, Adam, AdamW, RMSprop, Adagrad, RAdam, NAdam)
- 2k+ library tests in `flodl` (2.7k+ across the workspace), zero clippy warnings - the suite runs on CPU and on CUDA, with NCCL/DDP and serial gates of their own

### Hardware Compatibility

Developed and tested from NVIDIA Pascal (GTX 1060 6GB) to Blackwell
(RTX 5060 Ti 16GB). PyTorch dropped Pascal support after 2.5.1 - floDl
links libtorch's stable C API, which supports every architecture the driver
supports. If `nvidia-smi` works, floDl trains on it.

**AMD (ROCm)** builds, links and detects: `fdl libtorch download --rocm 7.0`
installs the ROCm build, `--features rocm` compiles against it, and GPUs are
enumerated from the kernel's KFD topology. Training on AMD is **not yet
validated on hardware** - the vendor axis is in place, the runtime numbers
are not. A libtorch build serves one vendor, so a host picks CUDA or ROCm,
never both in one process (and a join window refuses a cohort mixing the
two under NCCL/RCCL; the CPU averaging modes mix legally). Vendor-plural
today: detection (`fdl probe` / `diagnose`), variant install, the walk-in
path (`fdl join` / `fdl publish`), the container services and per-rank
device pinning. Still NVIDIA-only: fan-out's remote `local_devices: all`
probe, `fdl --gpus all` resolution, `fdl libtorch build` (source builds)
and `fdl nccl build` (RCCL ships inside libtorch-rocm).

Linux and [Windows via WSL2](https://github.com/flodl-labs/flodl/blob/main/docs/windows-wsl.md)
are the trained-on platforms - the benchmarks above were produced under
Docker on WSL2, so it is a measured path rather than a claimed one. Apple
Silicon runs CPU-only through Docker
([guide](https://github.com/flodl-labs/flodl/blob/main/docs/mac-apple-silicon.md)).

## Documentation

### Choose your path

| Background | Start here |
|-----------|-----------|
| **New to Rust** | [Rust for PyTorch Users](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/00-rust-primer.md) - 10 patterns in 15 minutes |
| **Know Rust, new to DL** | [Tensors](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/01-tensors.md) then [Training](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/04-training.md) |
| **Know PyTorch** | [Porting Guide](https://github.com/flodl-labs/flodl/blob/main/docs/porting.md) (or `/port` with AI) then [Graph Builder](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/05-graph-builder.md) |
| **Scaling to multi-GPU** | [Multi-GPU Training](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/11-multi-gpu.md) then [Heterogeneous & Multi-Host DDP](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/12-async-ddp.md) |
| **Bringing a HuggingFace model** | [HuggingFace Integration](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/14-flodl-hf.md): load BERT, RoBERTa, DistilBERT, ALBERT, XLM-R, or DeBERTa-v2; classify, NER, QA, or fill-mask; fine-tune and round-trip back to the HF ecosystem |
| **Just show me code** | [`quickstart`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/quickstart/) or [`showcase`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/showcase/) |
| **Forking or hacking on flodl itself** | [CONTRIBUTING.md](https://github.com/flodl-labs/flodl/blob/main/CONTRIBUTING.md) - clone-to-`fdl test` in four commands, the `*.example` templates you copy locally, and the gates a PR has to pass |

### Tutorials

0. **[Rust for PyTorch Users](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/00-rust-primer.md)** - 10 Rust patterns in 15 minutes
1. **[Tensors](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/01-tensors.md)** - creation, ops, memory, CUDA
2. **[Autograd](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/02-autograd.md)** - variables, gradients, backward
3. **[Modules](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/03-modules.md)** - all layers, convolutions, RNNs, attention, normalization
4. **[Training](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/04-training.md)** - losses, optimizers, mixed precision, full loop
5. **[Graph Builder](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/05-graph-builder.md)** - fluent API from simple to complex
6. **[Advanced Graphs](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/06-advanced-graphs.md)** - forward refs, loops, gates, switches
7. **[Visualization](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/07-visualization.md)** - DOT/SVG, profiling heatmaps
8. **[Utilities](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/08-utilities.md)** - checkpoints, clipping, freezing, initialization, scheduling, verbosity-gated logging
9. **[Training Monitor](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/09-monitor.md)** - ETA, resource tracking, live dashboard
10. **[Graph Tree](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/10-graph-tree.md)** - hierarchical composition, freeze/thaw, subgraph checkpoints
11. **[Multi-GPU Training](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/11-multi-gpu.md)** - Trainer::run / Trainer::builder, process-per-rank auto-promote, ElChe, DataLoader integration
12. **[Heterogeneous & Multi-Host DDP](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/12-async-ddp.md)** - ElChe cadence, process-per-rank cluster, A/B testable backends
13. **[Data Loading](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/13-data-loading.md)** - DataLoader, resident/streaming modes, VRAM-aware prefetch, DDP integration
14. **[HuggingFace Integration](https://github.com/flodl-labs/flodl/blob/main/docs/tutorials/14-flodl-hf.md)** - load BERT, RoBERTa, DistilBERT, ALBERT, XLM-RoBERTa, DeBERTa-v2 checkpoints, AutoModel dispatch across four task heads (seqcls, NER, QA, MLM), fine-tune with `Trainer::run` / `Trainer::builder(...)`, round-trip export to the HF ecosystem, PyTorch parity

### Examples

- [`quickstart`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/quickstart/): build, train, and monitor a model with residual connections
- [`sine_wave`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/sine_wave/): sine regression with monitor, checkpoint round-trip
- [`regression`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/regression/): linear and logistic regression as `Linear` plus the right loss
- [`mixed_precision`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/mixed_precision/): float16 training with `GradScaler`
- [`transfer_learning`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/transfer_learning/): checkpoint, partial load, freeze, fine-tune
- [`schedulers`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/schedulers/): warmup + cosine + plateau composition
- [`observation`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/observation/): collect, flush, trend queries, early stopping
- [`showcase`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/showcase/): every graph builder method in one graph
- [`flowbuilder_residual`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/flowbuilder_residual/): MLP classifier with a tagged residual block via FlowBuilder
- [`auto_promote`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/auto_promote/): the same plain training code auto-scaling CPU → 1 GPU → N-GPU DDP, zero distributed code
- [`bio/dnaseq_conv1d`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/bio/dnaseq_conv1d/): DNA Conv1D motif classifier (synthetic ChIP-seq peak annotation)
- [`bio/dna_autoencoder`](https://github.com/flodl-labs/flodl/tree/main/flodl/examples/bio/dna_autoencoder/): DNA convolutional autoencoder with checkpoint round-trip

### Porting from PyTorch

- **[Porting Guide](https://github.com/flodl-labs/flodl/blob/main/docs/porting.md)** - module mapping, FlowBuilder patterns, training loop translation
- **[AI-assisted porting](https://github.com/flodl-labs/flodl/tree/main/ai/skills/port/)** - point any AI coding assistant at the skill guide for automated translation. With Claude Code: `/port my_model.py`
- **`fdl api-ref`** - generate a structured API reference for your flodl version. Used by AI tools and useful on its own.

### Architecture

```
+-----------------------------------------------------------+
|  User Code / Model Definitions                            |
+-----------------------------------------------------------+
|  monitor/  ETA, resources, recursive dashboard portal     |
+-----------------------------------------------------------+
|  graph/    Fluent builder, graph tree, execution, DOT/SVG |
+-----------------------------------------------------------+
|  data/     DataLoader, resident/streaming, prefetch       |
+-----------------------------------------------------------+
|  distributed/  Trainer tiers, ElChe, cluster, NCCL        |
+-----------------------------------------------------------+
|  nn/       Modules, losses, optimizers, schedulers        |
+-----------------------------------------------------------+
|  autograd/ Reverse-mode AD, gradient tracking             |
+-----------------------------------------------------------+
|  tensor/   Owned tensors with Drop, CPU + CUDA            |
+-----------------------------------------------------------+
|  flodl-sys   FFI bindings to libtorch C++ shim            |
+-----------------------------------------------------------+
|  libtorch / CUDA / NCCL                                   |
+-----------------------------------------------------------+
```

## Story

floDl started as a question: what would a deep learning framework look like
if you designed it around Rust's ownership model instead of fighting a garbage
collector?

An [earlier attempt in Go](https://github.com/fab2s/goDl) proved the
architecture - the graph builder, the module system, the observation engine -
but hit the GC wall described above. Rust solved it at the language level, so
the graph builder, module composition, and design philosophy carried forward;
the memory fights didn't.

## License

floDl is open-sourced software licensed under the [MIT license](https://github.com/flodl-labs/flodl/blob/main/LICENSE).

flodl

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README