FireRedASR-AED Model for vLLM
This model has been converted for use with vLLM.
Architecture
- Encoder: Conformer (16 layers)
- Decoder: Transformer (16 layers)
- Model dimension: 1280
- Attention heads: 20
- Vocabulary size: 7832
- Precision: BFLOAT16
Tokenizer
✓ Tokenizer 已从 FireRedASR dict.txt 转换并包含在模型目录中。
Usage
This model is designed for automatic speech recognition tasks.
Special Token IDs
- BOS (Start): 3
- EOS (End): 4
- PAD (Padding): 2
