- 用 CharTokenizer.build_vocab 从 977K 数据生成词表 (vocab_size=47725) - train_cnb.py: batch=224 + accum=2 (等效448) + AMP + torch.compile - build_vocab.py: 词表生成脚本 - vocab_full.json: 含特殊token完整词表 - 模型 41.85M参数, L40 GPU训练
4 lines
131 BLFS
JSON
4 lines
131 BLFS
JSON
version https://git-lfs.github.com/spec/v1
|
|
oid sha256:c66bc101096011732fcf3ce3f033a7478fe3f9d8e98b5e64bb6f3e03f0d82b85
|
|
size 999336
|