Replace TrainingAdapter internals with ModelRuntime

Add GPU telemetry to benchmark_job.py

Add GPU telemetry to publish_job.py

Add GPU telemetry to eval_job.py

Allow planned sources in EvalJob dry-run

Allow planned sources in BenchJob dry-run

Allow planned sources in PublishJob dry-run

Add TrainingJob evaluation runtime

Add TrainingJob ONNX benchmark runtime

Fix PublishJob dependency resolution

Add TrainingJob ONNX publisher

Add TrainingJob checkpoint resume support

Support TrainingJob checkpoint resume

Fix eval_job.py dry-run compatibility

Fix publish_job.py dry-run compatibility

Add unified publish_job.py

Add unified eval_job.py

Add unified benchmark_job.py

Add checkpoint import and ONNX publish pipelines

Add checkpoint import and ONNX publish pipelines

Keep per-layer metadata valid when scaling child models

Add one-click distillation pipeline Job

Add compact child creation Job

Add live CPU RAM GPU VRAM telemetry

Use repository-native Unlimited OCR Nano training backend

Add Unlimited OCR Nano TrainingAdapter

Publish generic TrainingAdapter jobs

initial commit