From the model card
Fields below are copied from the tags and counters on the HuggingFace repository prajjwal1/bert-medium at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- prajjwal1
- Library
- Transformers
- Framework tags
- PyTorch
- License tag
mit— read the license file in the repo before relying on it- Language tags
- English (en)
- Papers cited
- arXiv:1908.08962, arXiv:2110.01518
- Downloads (HF counter at last fetch)
- 376,244
- Likes (HF counter at last fetch)
- 5
- Model card
- https://huggingface.co/prajjwal1/bert-medium
Use cases
- Medium-throughput text classification where BERT-base is too slow
- Named entity recognition on CPU-constrained infrastructure
- Embedding backbone for applications with strict latency SLAs
- Educational baseline for comparing BERT scale variants
Pros
- Good balance between BERT-small (too weak) and BERT-base (too slow)
- Widely used as a research baseline in efficiency-focused NLP studies
- Apache 2.0 license
- Loads quickly and fits in minimal memory
Cons
- Outperformed by DistilBERT on many tasks despite similar parameter count
- Not updated for DeBERTa-v3 or modern tokenization improvements
- Accuracy on complex NLP tasks degrades noticeably vs BERT-base
- Task-specific fine-tuning is still required for any real application
Tags
transformerspytorchBERTMNLINLItransformerpre-trainingenarxiv:1908.08962arxiv:2110.01518license:mitendpoints_compatibleregion:us