results.json

{
  "model": "AXL-Code-1B",
  "params": 317590528,
  "steps": 1012,
  "time": 1801.6977367401123,
  "final_loss": 2.939126491546631,
  "perplexity": 31.22,
  "gguf_size_mb": 606.3,
  "context_window": 256,
  "note": "Trained with vanilla SGD (no momentum) due to RAM constraints. Original target 978M params, trained at 318M."
}