Eager vs. JIT | CUDA | Writing GPU-friendly code | Data types | BLAS / LAPACK | Compilation cost | Padding inputs to avoid recompilation | Asynchronous execution | Memory | Donation | Eager-mode subset-assignment always copies

Adding a Primitive: Practical Example | Step 1: Define the Primitive | Special case: primitives with 0 dynamic inputs | Shortcut helpers for common shapes | Step 2: Add the StableHLO Rule | Step 3: Add the Reverse Rule | Optional: alternative-forward rule | Optional: a quickr rule | Step 4: Verify the Registration | Step 5: Add an nv_ API Function | Using Your Primitive | Contributing to the Package | File Organization | Testing | Higher-Order Primitives

Kernel | Joint Distribution | Posterior Predictive Distribution | Marginal Likelihood | Optimization | Results | Convergence with More Data | Further Reading

Problem | Data | Model | Training

The Banana Distribution | Metropolis-Hastings Algorithm | Running the Sampler | Inspecting the Results

Subsetting 1D arrays | Subsetting higher-dimensional arrays | Out-of-bounds Handling | Updating Subsets | Duplicate Indices | Copying Behavior

Pure Functions | Consistent Input and Output Types | Arbitrary Devices | Static Arguments to Enable Input Checks

Why is my function slow? | Why does timing my function show suspiciously fast results? | How do I control the number of threads used by XLA?

Transforming Code | Tracing R Functions into Graphs | Transforming Graphs into other Graphs | Lowering a Graph | The User Interface | jit() | gradient() | More Internals | Constant Handling | Device Inference in jit() | Nested Inputs and Outputs | Dichotomy of anvl functions

How jit() works | The compilation cache | Tracing | The Tracing Contract | R loops are unrolled | R if statements pick one branch | Closed-over values become constants | Side effects only fire during tracing | Other jit() arguments | Donating inputs | Device placement

The AnvlArray | Transforming AnvlArrays | Just In Time Compilation | Automatic Differentiation (AD) | Control Flow

System Dependencies | CPU Installation | GPU Installation | Docker | Available Images | Tags

The masking pattern | Masked sum | Masked mean | Subset assignment

Type Promotion Rules | Literals as Ambiguous Types | Creating Tensors with Different Ambiguity | Propagating Ambiguity