config.json

{
    "architectures": [
        "Lake1ForAnyToAny"
    ],
    "vocab_size": 20064,
    "hidden_size": 1024,
    "num_hidden_layers": 16,
    "num_attention_heads": 64,
    "embd_pdrop": 0.1,
    "use_moe": true,
    "num_experts": 24,
    "expert_hidden_size": 496,
    "dropout": 0.1,
    "model_type": "lake",
    "torch_dtype": "float16"
}