training_config.json

{
  "model_name": "Quantum-Clasifier",
  "task": "safe_vs_unsafe",
  "datasets": [
    "SetFit/enron_spam",
    "ucirvine/sms_spam",
    "deepset/prompt-injections"
  ],
  "label_semantics": {
    "0": "safe",
    "1": "unsafe"
  },
  "architecture": {
    "vocab_size": 4096,
    "context_length": 128,
    "d_model": 112,
    "attention_heads": 7,
    "layers": 4,
    "ffn": 336,
    "classifier_hidden": 170,
    "dropout": 0.1
  },
  "parameters": 1000000,
  "batch_size": 128,
  "target_tokens": 50000000,
  "actual_tokens": 50005046,
  "epochs": 17,
  "optimizer": "AdamW",
  "learning_rate": 0.0003,
  "minimum_learning_rate": 3e-05,
  "weight_decay": 0.01,
  "warmup_ratio": 0.05,
  "gradient_clip": 1.0,
  "mixed_precision": "fp16",
  "torch_compile": true,
  "confidence_temperature": 2.6337928771972656
}