reactantnitro-jl

Plugin reactantnitro-jl from marketplace reactantnitro-jl in EnzymeAD/ReactantNitro.jl. Commands read from the repo's manifest — 8 days ago.

Skills (11)

• reactantnitro-accelerators

  Choosing and configuring the accelerator a ReactantNitro run executes on: CPU, CUDA, ROCm, TPU, and how many devices it shards over. The one-process-one-XLA model, `setup_devices!` (REPL) and its tool form `nitro_setup` (Kaimon) as one shared code path, the `n_devs` pin and its precedence, and `CUDA

• reactantnitro-checkpoint-resume

  Checkpoint and resume a ReactantNitro run: `Checkpointer(; keep)` and its selectors (`Best`, `Last`, `Epochs`, `Every`; retention is their union), `scope = :run | :dir`, current-metric-only selection, the deprecated `TopKCheckpointer`, the `run => checkpoint` source grammar shared by `weights` and `

• reactantnitro-device-boundary

  Debug and prevent CPU/GPU divergence in ReactantNitro: why a fully green CPU test suite can pass while code has never once worked on a device, what a model author must and must not do at the boundary, what scalar indexing does differently on each backend, how to read a residency assertion, and the r

• reactantnitro-experiment

  Author or read a ReactantNitro experiment: the four required hooks, the three field markers (Device, Host, GraphConst) and which one a field wants, why every knob is a FLAT field and never a configuration struct held in one, the batch contract and keyword routing, what a hook may return and why a ba

• reactantnitro-export

  Export a trained ReactantNitro model to a servable artifact: the six export hooks and which are optional, why the exported graph is not the graph that trained, the wire convention, what the framework derives and verifies versus what you declare, and what provenance it stamps versus what you must sup

• reactantnitro-kaimon

  Drive ReactantNitro runs from a Kaimon-hosted session: the `nitro_*` GateTools the KaimonGate extension registers (train, validate, evaluate, predict, export, runs, status, stop), when the tools appear, why every tool returns immediately and the run runs in the background, how to name an experiment

• reactantnitro-manual

  Train a ReactantNitro experiment in manual mode, where the experiment owns the whole optimizer step: `train_step` and how defining it selects the mode, `setup_optimizers`, the checked return contract, `backward` with the objective discipline (a closure capture of a traced value silently zeroes the g

• reactantnitro-metrics

  Write metrics for a ReactantNitro experiment: the `value => mode` contract (a count, :sum, :max, :min, :concat) and why the framework does not supply the denominator, groups that share or mix modes, the fixed key set and why an empty stratum must still report, `finalize_metrics` for anything that is

• reactantnitro-optimizer

  Configure a ReactantNitro optimizer: parameter groups and per-group learning-rate ratios, `Decay` with its `:zero` and `:w0` anchors (L2 and L2-SP), per-leaf decay exclusion through `no_decay` and `default_no_decay`, global-norm gradient clipping, SAM through `grad_rule`, and schedules with their tw

• reactantnitro-recompiles

  Verify a ReactantNitro run is not recompiling, and work out why an edit did not take effect. Covers `cache_stats()` as the acceptance check, what is and is not in the compile key, why a changed GraphConst misses and a changed Device hits, the rule that a `Nitro` is a fixed point (so a revised hook n

• reactantnitro-visualization

  Build visualizations for a model on ReactantNitro: what to draw that a metric cannot tell you, why the failures rather than a random sample, the two hooks (`visualize`, `save_figure`) and the `render` driver that owns the iteration, the slicing, and the naming, the four moments worth rendering at, a

Similar plugins