Expand description
T5 model implementation with quantization support.
T5 is an encoder-decoder model pre-trained on a multi-task mixture of supervised and unsupervised tasks. This implementation provides quantization for reduced memory and compute requirements.
Key characteristics:
- Encoder-decoder architecture
- Layer normalization
- Relative positional encodings
- Support for 8-bit quantization
References:
- 📝 T5 Paper
- 🤗 Model Card
- 🤗 Original model from T5
Re-exports§
pub use crate::quantized_var_builder::VarBuilder;