evaluation.json

{
  "checkpoint": "gpu-additional-100m/best.pt",
  "checkpoint_step": 23000,
  "training_tokens": 148910080,
  "method": "token-weighted next-token cross-entropy over nonoverlapping context windows",
  "validation": {
    "tokens": 105075,
    "loss": 1.780266351765631,
    "perplexity": 5.931436056678534
  },
  "test": {
    "tokens": 102850,
    "loss": 1.8464635412079882,
    "perplexity": 6.3373680057668
  }
}