talos-10m / config.json
PrometheanStudio's picture
Release Talos 10M — 100k steps
d2d10c8 verified
Raw History Blame Contribute Delete
440 Bytes
{
"model": "Talos 10M",
"architecture": "TalosGPT",
"parameters": 9952320,
"vocab_size": 1024,
"training_tokenizer_vocab_size": 512,
"hidden_size": 448,
"num_layers": 3,
"num_attention_heads": 28,
"num_kv_heads": 14,
"head_dim": 16,
"intermediate_size": 1792,
"max_seq_len": 512,
"attention_type": "full",
"rope_theta": 10000,
"training_steps": 100000,
"validation_loss": 1.1523912596702575,
"seed": 1337
}