From the model card
Fields below are copied from the tags and counters on the HuggingFace repository google-t5/t5-small 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)
- google-t5
- Pipeline tag
- translation
- Library
- Transformers
- Framework tags
- PyTorch, TensorFlow, JAX, Rust (candle)
- Weight formats
- ONNX, safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Language tags
- multilingual; English (en), French (fr), Romanian (ro), German (de)
- Papers cited
- arXiv:1805.12471, arXiv:1708.00055, arXiv:1704.05426, arXiv:1606.05250, arXiv:1808.09121, arXiv:1810.12885, arXiv:1905.10044, arXiv:1910.09700
- Datasets declared
- c4
- Downloads (HF counter at last fetch)
- 22,647,932
- Likes (HF counter at last fetch)
- 597
- Model card
- https://huggingface.co/google-t5/t5-small
Use cases
- Teaching and experimenting with seq2seq architectures
- Fast baseline for summarization or translation research
- Lightweight fine-tuning when data is scarce
- Legacy pipeline compatibility where T5 is already deployed
Pros
- Unified text-to-text interface handles any NLP task
- Apache-2.0 licensed
- Lightweight at 60M parameters — fast CPU inference
- Extensive documentation and research literature
Cons
- Flan-T5 and mT5 outperform it with better instruction tuning
- 60M parameters produce low-quality output on generative tasks
- Outdated tokenizer and model architecture by current standards
- No chat or instruction-following capability without significant fine-tuning
Tags
transformerspytorchtfjaxrustonnxsafetensorst5text2text-generationsummarizationtranslationenfrrodemultilingualdataset:c4arxiv:1805.12471arxiv:1708.00055arxiv:1704.05426