AHA-L2A-Qwen3-1.7B-repro/recipe/source_training/train_results.json

{
    "ce_loss": 0.8887,
    "total_flos": 6.189458802853544e+18,
    "train_loss": 0.9857371548739723,
    "train_ppl": 2.431889,
    "train_runtime": 23521.9665,
    "train_samples_per_second": 37.607,
    "train_steps_per_second": 0.294
}