lucid-dl/roformer-chinese-base-mlm

RoFormer (Chinese base, Masked-LM)

https://arxiv.org/abs/2104.09864

Lucid port of transformers/junnyu/roformer_chinese_base, converted to Lucid-native safetensors.

Available weights

Tag	Params	GFLOPs	Size	Source
CLUECORPUSSMALL (default)	124.2M	—	622.16 MB	transformers

Usage

import lucid
import lucid.models as models
from lucid.models.weights import RoFormerMLMWeights

# default tag
model = models.roformer_mlm(pretrained=True)

# explicit tag (enum or string)
model = models.roformer_mlm(weights=RoFormerMLMWeights.CLUECORPUSSMALL)
model = models.roformer_mlm(pretrained="CLUECORPUSSMALL")

# feed token ids (tokenize with the matching lucid.utils.tokenizer)
input_ids = lucid.tensor([[101, 7592, 2088, 102]], dtype=lucid.int64)
out = model(input_ids)
logits = out.logits  # (B, T, vocab_size)

Conversion

Converted from transformers/junnyu/roformer_chinese_base via python -m tools.convert_weights roformer_mlm --tag CLUECORPUSSMALL. Key mapping + numerical parity verified against the source.

License

apache-2.0 — inherited from the original weights.

Citation

Su et al., "RoFormer: Enhanced Transformer with Rotary Position Embedding", 2021.