| defaults: | |
| - _base_fst_bert | |
| - _self_ | |
| # === Model Identity === | |
| model_name: fst_353M_bert | |
| # === Model Architecture === | |
| hidden_size: 1024 | |
| num_hidden_layers: 24 | |
| num_attention_heads: 16 | |
| intermediate_size: 4096 | |
| max_position_embeddings: 1024 | |
| max_seq_len: 1024 |