reproducibility/training.json

{
  "steps": 100,
  "gpu": "NVIDIA L4",
  "torch": "2.8.0+cu128",
  "cuda": "12.8",
  "python": "3.12.10",
  "selection": {
    "step": 75,
    "score": 1.0,
    "metric": "macro-f1",
    "test_used": false
  },
  "base_model": "vllm-sr/Vela-1.0-Encoder-307M",
  "base_revision": "720ab37904ce15054068d429381bfae7549f00f6",
  "method": "full; fresh head; final two encoder layers",
  "batch_size": 2,
  "accumulation": 2,
  "max_length": 256,
  "seed": 42,
  "learning_rate": 2e-05,
  "head_learning_rate": 0.0001,
  "dataset_files": {
    "train.jsonl": "69022836ee506922a74b3583088a242fcbedc5fd181c4dd7bcea4e7eca60f814",
    "dev.jsonl": "5909fe10961b0c6500a5e7d6ac9efd99b114c16f3545d45cd04036a070dbd1c3"
  }
}