training.json

{
  "schema_version": 1,
  "completed_unix_seconds": 1785688375,
  "dataset": "C:/tmp/dinovision-data/manifests/imagenette2-320.json",
  "dataset_manifest_sha256": "2071f89b2a7b077d7729641fc858e1225d057d11a797d16198c0d195d989bb89",
  "requested_images": 2500,
  "training_images": 2500,
  "model": "C:/tmp/dinovision-assets/model.safetensors",
  "model_sha256": "4610ad75edef83e75afdebf162d148dc628045ea6cbb83d67d4708c709c4f91d",
  "initialization": "random",
  "initialization_sha256": null,
  "seed": 0,
  "image_size": 224,
  "encoder_layers": 3,
  "batch_size": 8,
  "steps": 12000,
  "data_order": "seeded Fisher-Yates; reshuffled after every complete pass",
  "objective": "mean absolute error (L1)",
  "optimizer": "Adam(beta1=0.9,beta2=0.999,epsilon=1e-8), linear decay to a 0.0001 floor at 95% of updates",
  "initial_learning_rate": 0.002,
  "final_learning_rate": 0.000100000005,
  "decoder_parameters": 2012547,
  "encoding_seconds": 43.9207732,
  "training_seconds": 2303.373087,
  "final_l1": 0.034826964,
  "decoder": "C:/tmp/dinovision-runs-main-7561a64-v1/3l224-seed0/decoder.bin",
  "decoder_sha256": "dcc2b9560fbe0d59869ea2e37d900c1b72dbf01631b5869fd0bae6ca16dff846",
  "diagnostic_samples": "C:/tmp/dinovision-runs-main-7561a64-v1/3l224-seed0/decoder_samples.png"
}