Mini-Transformer / configs /model /medium_model.yaml
AlaBoussoffara's picture
organized code and set up chainlit for demos
2d52135
Raw
History Blame Contribute Delete
334 Bytes
name: medium_model
best_checkpoint_path: ./outputs/checkpoints/medium_model/best.ckpt
latest_checkpoint_path: ./outputs/checkpoints/medium_model/latest.ckpt
tokenizer: bpe_16k
d_model: 1024
num_heads: 16
num_layers: 8
d_ff: 2048
dropout_rate: 0.1
vocab_size: 16384
max_seq_len: 128
pad_id: 0
bos_id: 1
eos_id: 2
layer_norm_style: pre