Class TrainingReport
- Namespace
- LMKit.Finetuning
- Assembly
- LM-Kit.NET.dll
Machine-readable summary of a completed training run, exposed by LastReport after TrainToAdapter(string) returns. Every value is measured from the run itself; nothing is estimated, so the report alone is sufficient evidence for the three failure classes that dominate training support: the wrong device, a memory ceiling, and a throughput regression.
public sealed class TrainingReport- Inheritance
-
TrainingReport
- Inherited Members
Properties
- AdapterPath
Path the trained adapter was saved to.
- CompletedUtc
UTC timestamp the run finished.
- ComputePeakBytes
Standing device compute-buffer bytes after the training evals settled their peak, summed across GPU backends. The number behind the "rides the memory ceiling" log warning, now readable.
- ContextWindow
Context window the run resolved (tokens).
- DeviceDescription
Driver-reported description of that device, or "cpu".
- DeviceNumber
Device index the training model resided on, -1 for CPU.
- EpochsPlanned
Epochs the run was configured for.
- FinalAccuracy
Last training next-token accuracy, in [0, 1].
- FinalLoss
Last training loss the optimizer reported.
- FirstLoss
First training loss the optimizer reported.
- LearningRate
Learning rate the run started from.
- MeanStepSeconds
Mean wall-clock seconds per optimizer step.
- PackedSamples
Samples packed per training row (1 = no packing).
- Rank
LoRA rank the run trained with.
- SampleCount
Training samples the dataset contributed.
- Seed
Seed the run trained with (0 = nondeterministic draw).
- StepsCompleted
Optimizer steps completed, training passes only.
- TotalWallSeconds
Total wall-clock seconds from session start to adapter save.
- ValidationLoss
Last validation loss, or NaN when no validation split ran.