{ "architecture": "qwen2", "epoch": 5, "evaluation_backend": "causal", "global_step": 11160, "jieba": true, "source_checkpoint": "/lustre/nec/ws3/ws/hpctrobr-babylm/models/babylm-qwen2-33m-bpe/best.pt", "tokenizer_kind": "sentencepiece", "transliteration": "pinyin-code", "use_jieba": true, "validation_loss": 5.256962834286089 }