{ "d_model": 1024, "n_heads": 8, "n_kv_heads": 4, "head_dim": 128, "n_layers": 18, "vocab_size": 262144, "ffn_hidden": 4096, "chunk_size": 8, "max_cache_chunks": 64, "n_modalities": 4, "n_tasks": 6, "n_languages": 4022, "n_mtp_heads": 4 }