# rlx-tensor
A native [`ndarray`](https://crates.io/crates/ndarray) alternative for RLX:
NumPy-style tensors with operator-overloaded, expression-like syntax that
**trace into rlx-ir instead of executing eagerly**. The graph stays lazy —
so the compiler fuses the whole expression and plans memory across every
backend — until you force it.
| Default execution | eager, host | eager, GPU | trace + XLA | **trace → fuse → backend** |
| Intermediate buffers | always materialized | often kept for autograd | XLA decides | **fusion + memory plan** |
| Device | host only | CUDA-centric | TPU / GPU | **CPU / Metal / MLX / CUDA / ROCm / wgpu / TPU** |
| Variable batch / seq | manual padding | manual | `vmap` / shapes | **first-class `Dim::Dynamic`** |
| Slicing | views (cheap) | views | slice ops | **`s![]` → `narrow` (fusable)** |
The base crate is a pure rlx-ir graph builder — symbolic `Tensor` handles carry
**zero payload bytes** and pull in **no backend**. Materialization, autodiff,
and training are opt-in features so you only pay for what you use.
## Quickstart
```rust
use rlx_tensor::Tensor; // requires feature = "eval"
let a = Tensor::from_vec(vec![1.0, 2.0, 3.0], [3]);
let b = Tensor::ones([3]);
let c = (&a + &b).relu(); // graphs auto-merge, nothing runs yet
assert_eq!(c.to_vec(), vec![2.0, 3.0, 4.0]); // compiles once, auto-picks the fastest backend
// c.on(Device::Metal).to_vec() // or pin an explicit backend
```
Materialization is compile-cached per thread (keyed by graph + constants +
device), so repeated `to_vec` / `Func::run` reuse the compiled artifact.
## Building a graph
Two equivalent styles — eager-looking `Tensor` constructors (need `eval` to
read back), or a pure symbolic `graph(..)` scope with named inputs/params:
```rust
use rlx_tensor::{ax, graph, rg, s, shape};
// static shape
let w = g.param("w", shape![4, 3]);
x.matmul(&w).gelu()
});
// dynamic batch + window slice (no copy in IR)
x.slice(s![ax(), rg(0, 64)]) // s![] → narrow view
});
```
## Op surface
- **Arithmetic** — `+ - * /` (operator overloads, tensor & scalar rhs),
`maximum` / `minimum` / `clamp` / `pow`.
- **Activations** — `relu`, `gelu`, `gelu_approx`, `silu`, `sigmoid`, `tanh`,
`exp`, `log`, `sqrt`, `rsqrt`, `abs`, `sin`, `cos`, `tan`, `atan`, `round`.
- **Reductions** — `sum`, `mean`, `max`, `min`, `prod`, `var`, `std`, `norm`,
`logsumexp`, `cumsum`, `argmax`, `argmin`.
- **Shape / view** — `reshape`, `transpose` / `t`, `narrow`, `slice`, `select`,
`split`, `chunk`, `flatten`, `squeeze` / `unsqueeze`, `broadcast_to`, `tile`,
`flip`, `roll`, `pad`, `cat`, `stack`.
- **Indexing** — `gather`, `index_select`, comparisons (`eq` `ne` `lt` `le`
`gt` `ge`), `where_`, `masked_fill`.
- **NN / linalg** — `matmul` / `mm`, `softmax`, `layer_norm`, `rms_norm`,
`conv2d`, `attention`, `rope`, `fft` / `ifft` / `fft_axis`, `inv`, `solve`.
- **Constructors** — `from_vec`, `zeros`, `ones`, `full`, `eye`, `arange`,
`arange_step`, `rand`, `randn`.
## Autodiff (feature `grad`)
Reverse-mode gradients on a materialized tensor:
```rust
let loss = (&a * &b).sum([0], false);
let g = loss.grad(&[&a, &b]); // [∂loss/∂a, ∂loss/∂b]
```
## Composable transforms (feature `transforms`)
JAX-shaped transforms operate on `Func` (a traced function of named inputs) and
chain — `vmap(grad(f))`:
```rust
(&x * &x).sum([0], false)
});
let batched_grad = f.grad(&["x"]).vmap(&["x"], 4); // Func -> Func -> Func
let compiled = batched_grad.jit(); // compile once
let out = compiled.run(&[("x", &xs)]); // run many, no recompile
```
`Func` also exposes `value_and_grad`, `jvp`, and `hvp`.
## Training (feature `optim`)
`Func::train_step` runs value+grad and applies an `rlx_optim` optimizer to the
bound params:
```rust
use rlx_tensor::{AdamW, Func};
let mut opt = AdamW::new(3e-4);
let (loss, _) = f.train_step(&mut opt, &[("x", &xs)]);
```
`Adam`, `AdamW`, `Lion`, `Sgd`, `Optimizer`, and `LrSchedule` are re-exported.
## Features
| *(default)* | pure rlx-ir graph builder — symbolic only, no backend |
| `eval` | materialize via `rlx_runtime::Session`: `to_vec`, `on(device)`, CPU backend |
| `eval-metal` / `eval-mlx` / `eval-cuda` / `eval-rocm` / `eval-gpu` / `eval-coreml` | `.on(Device::…)` for that backend |
| `eval-apple` | all Apple GPU/NPU backends (Metal + MLX + wgpu + CoreML/ANE) |
| `eval-blas` | CPU eval with Accelerate BLAS/LAPACK (needed for `inv` / `solve`) |
| `grad` / `transforms` / `autodiff` | reverse-mode AD + composable `Func` transforms (implies `eval` where it materializes) |
| `optim` | `Func::train_step` + `rlx_optim` optimizers (implies `autodiff` + `eval`) |
| `ndarray` | `Tensor::from(array)` / `tensor.to_ndarray()` interop with `ndarray` |
Migrating from `ndarray`? Enable `ndarray` for zero-friction `From`/`to_ndarray`
round-trips.
## Where it sits
`rlx-tensor` is re-exported through the prelude crate as
[`rlx::tensor`](../rlx) (`use rlx::prelude::*;` brings `Tensor`, `graph`, the
`s!` / `shape!` macros, and the slice helpers into scope). It is one layer above
[`rlx-ir`](../rlx-ir) and shares its `Op` set, so any graph it builds compiles
through the same fusion / memory-planning / backend pipeline as the rest of RLX.