rootxhacker/HobbyLM-Computer-Use
023
1{2 "vocab_size": 50304,3 "d_model": 768,4 "n_layers": 16,5 "n_dense_layers": 1,6 "n_q_heads": 12,7 "n_kv_heads": 3,8 "head_dim": 128,9 "qk_norm": true,10 "rope_theta": 10000.0,11 "attn_softcap": 0.0,12 "dense_ffn": 2304,13 "expert_ffn": 320,14 "n_experts": 36,15 "top_k": 6,16 "n_shared": 1,17 "gating": "sigmoid",18 "norm_topk_prob": false,19 "balancing": "aux_free",20 "aux_loss_coef": 0.001,21 "z_loss_coef": 0.001,22 "bias_update_rate": 0.001,23 "router_init_std": 0.02,24 "tie_embeddings": true,25 "scale_embeddings": false,26 "final_z_loss_coef": 0.0001,27 "logit_softcap": 0.0,28 "n_mtp": 0,29 "mtp_weight": 0.1,30 "init_std": 0.02,31 "expert_backend": "grouped",32 "fused_ce": true,33 "ce_chunk": 4096,34 "fp8_head": false,35 "fp8_x_scale": 1.0,36 "fp8_w_scale": 1.0,37 "fp8_grad_scale": 1.0,38 "preset": "500M"39}