deepj-tinystories / config.json
netsrik's picture
Use DeepJ Prism model type
ec859bb verified
Raw History Blame Contribute Delete
318 Bytes
{
"library_name": "deepj",
"model_type": "deepj-prism",
"checkpoint_format_version": 1,
"tokenizer_format_version": 1,
"vocab_size": 2048,
"max_seq_len": 128,
"d_model": 128,
"n_heads": 4,
"n_layers": 4,
"d_ff": 384,
"q_rank": 64,
"kv_rank": 32,
"init_scale": 0.2,
"grad_clip_norm": 1.0
}