Training Command-Line Options

Command-line options shared by runnable training programs. These records describe process-level choices such as step counts, log destinations, and CUDA allocator sampling. Model definitions and the Trainer.train API do not depend on them.

structure TorchLean.CLI.Training.RunOptions : Type

Step, batching, logging, and allocator-reporting options for a training command.

• steps : ℕ

  Number of optimizer updates.

• batchSize : ℕ

  Number of in-memory samples consumed by one optimizer update.

• log : Training.LogDestination

  Destination for the JSON training log.

• logPath : System.FilePath

  Resolved log path, retained for command summaries.

• cudaMemWatch : ℕ

  Number of completed steps between CUDA allocator samples; 0 selects the default policy.

@[implicit_reducible]

instance TorchLean.CLI.Training.instReprRunOptions : Repr RunOptions

def TorchLean.CLI.Training.instReprRunOptions.repr : RunOptions → ℕ → Std.Format

def TorchLean.CLI.Training.RunOptions.parse (exeName : String) (args : List String) (defaultLogPath : System.FilePath) (defaultSteps : ℕ := 1) (allowZeroSteps : Bool := false) : Except String (RunOptions × List String)

Parse the common options accepted by runnable training commands.

def TorchLean.CLI.Training.RunOptions.writeLossComparison (opts : RunOptions) (title : String) (beforeLoss afterLoss : Float) (notes : Array String := #[]) : IO Unit

Write a before-and-after loss log using the command's selected destination and step count.

structure TorchLean.CLI.Training.OptimizerOptionsextends TorchLean.CLI.Training.RunOptions : Type

Training command options that also select a learning rate.

• steps : ℕ
• batchSize : ℕ
• log : Training.LogDestination
• logPath : System.FilePath
• cudaMemWatch : ℕ
• lr : Float

  Learning rate passed to the command's optimizer constructor.

@[implicit_reducible]

instance TorchLean.CLI.Training.instReprOptimizerOptions : Repr OptimizerOptions

def TorchLean.CLI.Training.instReprOptimizerOptions.repr : OptimizerOptions → ℕ → Std.Format

def TorchLean.CLI.Training.OptimizerOptions.parse (exeName : String) (args : List String) (defaultLogPath : System.FilePath) (defaultSteps : ℕ := 1) (defaultLr : Float := 1e-3) (allowZeroSteps : Bool := false) : Except String (OptimizerOptions × List String)

Parse run options followed by a positive --lr value.

def TorchLean.CLI.Training.OptimizerOptions.writeLossComparison (opts : OptimizerOptions) (title : String) (beforeLoss afterLoss : Float) (notes : Array String := #[]) : IO Unit

Write a before-and-after loss log for an optimizer command.