lnp_ml/models/final/best_params.json

16 lines
394 B
JSON

{
"dropout": 0.19323529834582587,
"lr": 0.000980484886752915,
"weight_decay": 0.00014367509113739864,
"backbone_lr_ratio": 0.13093310334961422,
"moe_n_experts": 2,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 8,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}